WEBVTT
1
00:00:04.120 --> 00:00:07.240
Sarath Ranganathan: Whether you're a student,
staff member, alumni,
2
00:00:07.240 --> 00:00:10.520
Sarath Ranganathan: or someone who's passionate about healthcare
and its future,
3
00:00:10.920 --> 00:00:13.400
Sarath Ranganathan: you're listening to the right podcast.
4
00:00:21.560 --> 00:00:26.320
Sarath Ranganathan: This podcast was recorded on the lands of the
Wurundjeri people of the Kulin nation,
5
00:00:26.680 --> 00:00:30.760
Sarath Ranganathan: and we pay our respects to elders,
past and present.
6
00:00:36.320 --> 00:00:38.215
Sarath Ranganathan: Welcome to MMS Network.
7
00:00:38.215 --> 00:00:41.956
Sarath Ranganathan: I'm your host, Professor Sarah Ranganathan,
head of the Melbourne Medical School,
8
00:00:41.956 --> 00:00:46.680
Sarath Ranganathan: and today we'll be discussing one of the most
transformative forces in modern medicine,
9
00:00:47.160 --> 00:00:49.360
Sarath Ranganathan: that is artificial intelligence.
10
00:00:49.840 --> 00:00:52.720
Sarath Ranganathan: And joining me in the studio is Doctor
Emerson Keenan,
11
00:00:52.880 --> 00:00:58.240
Sarath Ranganathan: co-founder of Kali Healthcare and a research
fellow in our department of Obstetrics,
12
00:00:58.400 --> 00:01:00.690
Sarath Ranganathan: Gynecology and Newborn Health.
13
00:01:01.090 --> 00:01:05.010
Sarath Ranganathan: Emerson's work sits at the intersection of AI
and medical devices,
14
00:01:05.370 --> 00:01:09.530
Sarath Ranganathan: reshaping how we think about patient care and
clinical research.
15
00:01:10.410 --> 00:01:12.650
Sarath Ranganathan: Welcome, Emerson. It's great to have you
here.
16
00:01:13.370 --> 00:01:17.490
Sarath Ranganathan: To start us off, could you tell us a little
bit more about your background?
17
00:01:17.690 --> 00:01:21.210
Emerson Keenan: Fantastic. Thanks so much for having me,
Sarath. So I'm actually an electrical
18
00:01:21.210 --> 00:01:22.330
Emerson Keenan: engineer by background.
19
00:01:22.330 --> 00:01:27.010
Emerson Keenan: So I did my original undergraduate studies in
Brisbane in electrical engineering,
20
00:01:27.010 --> 00:01:31.290
Emerson Keenan: and my first taste of medicine was in my
honours year doing a biomedical project,
21
00:01:31.290 --> 00:01:34.770
Emerson Keenan: looking at a smartphone application for
detecting sleep apnea.
22
00:01:34.770 --> 00:01:36.250
Emerson Keenan: And so that was really exciting.
23
00:01:36.250 --> 00:01:39.850
Emerson Keenan: And they went on to be commercialized through
a company called ResApp. And so from there
24
00:01:39.890 --> 00:01:41.690
Emerson Keenan: kind of sparked my interest in medicine.
25
00:01:41.690 --> 00:01:46.370
Emerson Keenan: So from there, I began a PhD at the
University of Melbourne looking at new
26
00:01:46.370 --> 00:01:50.530
Emerson Keenan: technologies for pregnancy monitoring. And so
that evolved over time to now spin out into
27
00:01:50.570 --> 00:01:54.490
Emerson Keenan: Kali Healthcare. And so I also have an
ongoing appointment as an honorary research
28
00:01:54.490 --> 00:01:57.290
Emerson Keenan: fellow in the Melbourne Medical School,
looking at how we apply artificial
29
00:01:57.290 --> 00:02:00.220
Emerson Keenan: intelligence to things like wearables,
large data sets,
30
00:02:00.260 --> 00:02:02.580
Emerson Keenan: biomedical data, and it's a really exciting
space to be in.
31
00:02:02.820 --> 00:02:07.260
Sarath Ranganathan: So when you started electrical engineering,
did you have any idea then that this was the
32
00:02:07.260 --> 00:02:09.180
Sarath Ranganathan: journey you were going to find yourself on?
33
00:02:09.220 --> 00:02:13.020
Emerson Keenan: No, I think I think back then I was just
really excited by what technology can do and
34
00:02:13.020 --> 00:02:17.180
Emerson Keenan: the power of humans to use technology to
solve really important problems.
35
00:02:17.180 --> 00:02:19.540
Emerson Keenan: But over time, I found more and more that,
you know,
36
00:02:19.540 --> 00:02:21.700
Emerson Keenan: solving these problems is one of the most
impactful things I could do with that
37
00:02:21.700 --> 00:02:22.940
Emerson Keenan: electrical engineering skill.
38
00:02:23.140 --> 00:02:27.060
Sarath Ranganathan: So tell us a little bit more about what your
PhD focused on,
39
00:02:27.260 --> 00:02:30.220
Sarath Ranganathan: and also how it evolved into Kali Healthcare.
40
00:02:30.580 --> 00:02:35.020
Emerson Keenan: My PhD was originally looking at ways that we
can better monitor pregnancies,
41
00:02:35.060 --> 00:02:36.620
Emerson Keenan: ideally using wearable devices.
42
00:02:36.620 --> 00:02:39.060
Emerson Keenan: So the crux of it is that the current
technology,
43
00:02:39.060 --> 00:02:41.940
Emerson Keenan: which uses ultrasound,
is a really hospital centric approach.
44
00:02:42.100 --> 00:02:45.700
Emerson Keenan: So women who have a pregnancy complication
are coming into hospitals regularly to have
45
00:02:45.700 --> 00:02:48.100
Emerson Keenan: monitoring done. It's really time consuming
for women,
46
00:02:48.100 --> 00:02:51.540
Emerson Keenan: time consuming for clinicians. So we looked
at ways of technologies that could be used
47
00:02:51.540 --> 00:02:54.580
Emerson Keenan: outside of that clinical setting so that we
can free up clinician time.
48
00:02:54.780 --> 00:02:58.470
Emerson Keenan: And so throughout my PhD,
I found that electrical signals seem to be a
49
00:02:58.470 --> 00:03:02.230
Emerson Keenan: really promising way of doing this monitoring
without needing a clinician to be physically
50
00:03:02.230 --> 00:03:05.790
Emerson Keenan: present. So originally it was looking at
developing the sensor patch,
51
00:03:05.790 --> 00:03:08.710
Emerson Keenan: developing the layout and the algorithms to
process things.
52
00:03:08.710 --> 00:03:12.230
Emerson Keenan: And over time, we found that by putting these
sensors in the right place and the right
53
00:03:12.230 --> 00:03:16.270
Emerson Keenan: configuration with the right algorithms,
which eventually become an artificial
54
00:03:16.270 --> 00:03:19.230
Emerson Keenan: intelligence algorithm,
we could get really reliable results and
55
00:03:19.230 --> 00:03:21.390
Emerson Keenan: unlock this ability to do monitoring and home
safely.
56
00:03:21.710 --> 00:03:26.270
Sarath Ranganathan: So how does the device work and how do you
integrate that into clinical care for
57
00:03:26.310 --> 00:03:27.230
Sarath Ranganathan: monitoring women?
58
00:03:27.350 --> 00:03:32.350
Emerson Keenan: Yeah. So the device essentially is a smart
device which uses a wearable sensor patch.
59
00:03:32.470 --> 00:03:35.510
Emerson Keenan: So we have that wearable sensor that goes on
the mum's abdomen.
60
00:03:35.510 --> 00:03:39.590
Emerson Keenan: There's a small unit attached to it which has
all the data collection infrastructure.
61
00:03:39.590 --> 00:03:41.190
Emerson Keenan: So we can pick up those signals.
62
00:03:41.190 --> 00:03:44.670
Emerson Keenan: It then uses Bluetooth to send the
information to the mum's smartphone.
63
00:03:44.710 --> 00:03:48.270
Emerson Keenan: That smartphone helps guide them through the
process of putting on the device,
64
00:03:48.310 --> 00:03:51.470
Emerson Keenan: getting that measurement reading,
as well as then sending that information over
65
00:03:51.470 --> 00:03:55.350
Emerson Keenan: the internet back to the clinician where they
can do a review. So it's really that
66
00:03:55.350 --> 00:03:58.880
Emerson Keenan: combination of that, that wearable device,
a smartphone that helps guide the user
67
00:03:58.920 --> 00:04:02.600
Emerson Keenan: through the process, as well as that internet
enabled platform that makes this all work.
68
00:04:03.040 --> 00:04:08.840
Sarath Ranganathan: So what sort of decisions do the clinicians
make based on what data or advice the device
69
00:04:08.880 --> 00:04:09.920
Sarath Ranganathan: is sending to them?
70
00:04:10.240 --> 00:04:14.840
Emerson Keenan: The initial focus of our technology is to
unlock new cases where we can use this.
71
00:04:14.840 --> 00:04:17.960
Emerson Keenan: So having that remote monitoring is really
our first launch point.
72
00:04:17.960 --> 00:04:22.320
Emerson Keenan: So the ability to unlock this care at home
freeing up clinician time's really valuable.
73
00:04:22.320 --> 00:04:26.040
Emerson Keenan: So the idea is we're giving them the same
information they're used to seeing from that
74
00:04:26.040 --> 00:04:29.440
Emerson Keenan: clinical setting. So that's in the initial
instance the fetal heart rate,
75
00:04:29.480 --> 00:04:32.320
Emerson Keenan: the maternal heart rate and the measure of
the contractions.
76
00:04:32.320 --> 00:04:34.640
Emerson Keenan: And they can see that all in the patient
dashboard.
77
00:04:34.760 --> 00:04:38.680
Emerson Keenan: We're working over time to bring more things
into the platform. So looking at bringing in
78
00:04:38.680 --> 00:04:41.960
Emerson Keenan: things like continuous glucose monitoring,
blood pressure monitoring,
79
00:04:42.000 --> 00:04:43.920
Emerson Keenan: having this all part of that platform.
80
00:04:43.920 --> 00:04:48.760
Emerson Keenan: And then really the grand vision behind all
of this is to add clinical decision support
81
00:04:48.760 --> 00:04:52.000
Emerson Keenan: into it. So not only providing that data over
the internet,
82
00:04:52.000 --> 00:04:55.280
Emerson Keenan: but having the ability to predict things
before they eventuate.
83
00:04:55.280 --> 00:05:00.120
Emerson Keenan: So one of the first things we're looking at
is using the uterine contraction information
84
00:05:00.120 --> 00:05:04.000
Emerson Keenan: to better pick up when women are going to go
into labor and predicting if some people
85
00:05:04.000 --> 00:05:07.120
Emerson Keenan: have, particularly in early labor and maybe
have preterm birth.
86
00:05:07.360 --> 00:05:10.600
Emerson Keenan: So we've got some early evidence that shows
using AI techniques,
87
00:05:10.600 --> 00:05:13.600
Emerson Keenan: that we can predict these cases where women
are going into labor early,
88
00:05:13.640 --> 00:05:15.320
Emerson Keenan: up to seven days in advance.
89
00:05:15.320 --> 00:05:17.760
Emerson Keenan: So having something like that,
if a doctor's doing a remote monitoring
90
00:05:17.760 --> 00:05:20.160
Emerson Keenan: session for a complicated pregnancy and they
say,
91
00:05:20.160 --> 00:05:23.440
Emerson Keenan: hey, there's a really strong likelihood that
you're going to go into delivery, they can
92
00:05:23.440 --> 00:05:26.360
Emerson Keenan: better triage that care. You know,
they can bring them into hospital, monitor
93
00:05:26.360 --> 00:05:30.760
Emerson Keenan: them more closely, and really only bring the
women in who are at the highest risk into the
94
00:05:30.760 --> 00:05:34.320
Emerson Keenan: hospital setting, and everyone else can stay
at home safely. So we're really excited about
95
00:05:34.600 --> 00:05:37.680
Emerson Keenan: where this can go beyond just unlocking those
new models of care,
96
00:05:37.720 --> 00:05:39.640
Emerson Keenan: moving into that predictive analytics space.
97
00:05:40.160 --> 00:05:43.880
Sarath Ranganathan: So from the mother's point of view,
someone who's wearing the device,
98
00:05:43.880 --> 00:05:49.560
Sarath Ranganathan: what do they notice? Or what would they tell
you about the benefits of that device?
99
00:05:49.600 --> 00:05:53.560
Emerson Keenan: Yeah. So throughout my PhD,
I assisted with a lot of the data collection.
100
00:05:53.560 --> 00:05:56.770
Emerson Keenan: So I was there in the clinic,
helping set up the device,
101
00:05:56.770 --> 00:06:00.370
Emerson Keenan: talking to women about their experiences. And
it was really incredible to see,
102
00:06:00.410 --> 00:06:04.690
Emerson Keenan: you know, just the reoccurring theme of how
much impact it has on women's lives having to
103
00:06:04.730 --> 00:06:08.690
Emerson Keenan: come in for this monitoring repeatedly,
the anxiety it brings when they're unsure at
104
00:06:08.690 --> 00:06:11.690
Emerson Keenan: home, they're thinking,
oh, should I come in, should I stay? And I
105
00:06:11.690 --> 00:06:15.850
Emerson Keenan: think that's a really important thing that we
think we can solve with this technology.
106
00:06:15.850 --> 00:06:19.610
Emerson Keenan: And so we actually incorporated this into one
of our most recent research studies,
107
00:06:19.610 --> 00:06:23.210
Emerson Keenan: where we asked women about their experiences
in having fetal monitoring.
108
00:06:23.210 --> 00:06:26.690
Emerson Keenan: And we published this recently in Nature
Digital Medicine,
109
00:06:27.090 --> 00:06:29.850
Emerson Keenan: which found that over 90% of women were
saying,
110
00:06:29.890 --> 00:06:33.610
Emerson Keenan: you know, the ability of this to reduce that
anxiety and the desire for having remote
111
00:06:33.610 --> 00:06:35.090
Emerson Keenan: monitoring was really high.
112
00:06:35.410 --> 00:06:38.570
Emerson Keenan: And so we think that,
you know, that that lens from the user
113
00:06:38.570 --> 00:06:41.130
Emerson Keenan: perspective is really important. It's not
just, you know,
114
00:06:41.170 --> 00:06:44.690
Emerson Keenan: solving some of these clinical workflow
challenges. It's also solving for mums that
115
00:06:44.690 --> 00:06:48.330
Emerson Keenan: that anxiety that, you know,
burden on them having to come into the
116
00:06:48.330 --> 00:06:50.930
Emerson Keenan: clinic, and particularly women who've got
kids at home,
117
00:06:50.970 --> 00:06:54.260
Emerson Keenan: you know, finding childcare for the kids or
finding people who can take care of other
118
00:06:54.260 --> 00:06:56.260
Emerson Keenan: family members. You know,
freeing all that up.
119
00:06:56.260 --> 00:06:59.740
Emerson Keenan: I think you know that the human aspect of the
technology is really important, you know,
120
00:06:59.780 --> 00:07:03.980
Emerson Keenan: helping solve not only clinical challenges,
but human usability challenges as well.
121
00:07:04.420 --> 00:07:06.540
Sarath Ranganathan: Now, that sounds incredibly positive.
122
00:07:07.140 --> 00:07:10.740
Sarath Ranganathan: Do you mind me then asking why you called
this Kali Healthcare,
123
00:07:10.780 --> 00:07:13.220
Sarath Ranganathan: given what I know about the goddess Kali?
124
00:07:13.260 --> 00:07:16.940
Emerson Keenan: Yeah. So originally the company was founded
by myself,
125
00:07:17.060 --> 00:07:19.940
Emerson Keenan: my engineering supervisor,
Marimuthu Palaniswami,
126
00:07:20.140 --> 00:07:22.660
Emerson Keenan: and my clinical colleague,
Doctor Fiona Brownfoot,
127
00:07:22.700 --> 00:07:24.780
Emerson Keenan: who's also in the obstetrics department.
128
00:07:24.780 --> 00:07:30.980
Emerson Keenan: And so when we're talking about ideas behind
the naming of the company Professor Marimuthu
129
00:07:30.980 --> 00:07:32.860
Emerson Keenan: Palaniswami suggested the goddess Kali.
130
00:07:32.860 --> 00:07:36.620
Emerson Keenan: And in his region, it's seen as a goddess of
fertility and motherhood.
131
00:07:36.620 --> 00:07:39.420
Emerson Keenan: And so we thought there was a really fitting
thing to,
132
00:07:39.460 --> 00:07:42.500
Emerson Keenan: you know, put behind the technology and
having that protective element.
133
00:07:42.860 --> 00:07:47.020
Sarath Ranganathan: Thanks for the explanation. I obviously
didn't understand the full skill sets of the
134
00:07:47.060 --> 00:07:51.380
Sarath Ranganathan: goddess Kali. So it's not really related to
AI itself,
135
00:07:51.380 --> 00:07:56.150
Sarath Ranganathan: but I did want to come on to ask about what
specific role AI now plays in that
136
00:07:56.150 --> 00:07:59.230
Sarath Ranganathan: monitoring. How did you integrate that into
the tool that you have?
137
00:07:59.470 --> 00:08:01.870
Emerson Keenan: So there's two key aspects of how we're using
AI.
138
00:08:01.910 --> 00:08:05.870
Emerson Keenan: So the first is the underlying processing of
the sensor data we collect.
139
00:08:05.870 --> 00:08:09.190
Emerson Keenan: So the sensor patch collects electrical
signals.
140
00:08:09.190 --> 00:08:13.590
Emerson Keenan: So all it's measuring is how those electrical
signals rise up and down on the abdomen.
141
00:08:13.590 --> 00:08:16.350
Emerson Keenan: And so that signal contains a really large
mixture.
142
00:08:16.350 --> 00:08:19.230
Emerson Keenan: It's got everything from the mum's heart rate
activity,
143
00:08:19.270 --> 00:08:22.630
Emerson Keenan: the baby's heart rate,
the muscle contractions of the uterus,
144
00:08:23.110 --> 00:08:26.630
Emerson Keenan: the other muscle contractions as they might
be moving around. And so it's really
145
00:08:26.630 --> 00:08:29.870
Emerson Keenan: difficult to separate all this out and unlock
those different signals.
146
00:08:29.870 --> 00:08:34.070
Emerson Keenan: And so we tried early in my PhD kind of
traditional approaches to extract this
147
00:08:34.070 --> 00:08:37.550
Emerson Keenan: information. And we just couldn't quite get
the reliability that we needed. And so
148
00:08:37.870 --> 00:08:40.950
Emerson Keenan: looking at AI gave us that breakthrough we
were looking for.
149
00:08:40.950 --> 00:08:45.190
Emerson Keenan: So the real difference here is that in
traditional methods we we have an idea about
150
00:08:45.190 --> 00:08:48.550
Emerson Keenan: how we think we want to extract the signal.
And an engineer kind of crafts an algorithm
151
00:08:48.550 --> 00:08:51.720
Emerson Keenan: based on the expert information they have,
whereas AI,
152
00:08:51.720 --> 00:08:54.640
Emerson Keenan: instead flips it on its head and says,
okay, we've got a signal,
153
00:08:54.640 --> 00:08:57.200
Emerson Keenan: we've got as an input and a signal we want to
get to as an output,
154
00:08:57.200 --> 00:09:00.080
Emerson Keenan: and we let the machine do the learning. So
essentially,
155
00:09:00.080 --> 00:09:04.160
Emerson Keenan: we've collected a very large data set of
recordings where we had both our device and
156
00:09:04.160 --> 00:09:08.280
Emerson Keenan: another technology on generally the
ultrasound based technology and measuring
157
00:09:08.280 --> 00:09:12.160
Emerson Keenan: them together, we can develop a model that
takes that input electrical signal and gives
158
00:09:12.160 --> 00:09:15.000
Emerson Keenan: us that output fetal heart rate with a much
higher accuracy.
159
00:09:15.160 --> 00:09:17.160
Emerson Keenan: So that's the first way we're using things.
160
00:09:17.160 --> 00:09:19.960
Emerson Keenan: And then the second lens,
which I was talking about is the ability to
161
00:09:19.960 --> 00:09:21.200
Emerson Keenan: do predictive analytics.
162
00:09:21.200 --> 00:09:24.680
Emerson Keenan: So not only can we predict those raw
physiological parameters,
163
00:09:24.680 --> 00:09:28.960
Emerson Keenan: but looking ahead to things like the
likelihood that women might go into labor.
164
00:09:28.960 --> 00:09:33.520
Emerson Keenan: And so there's things like the uterine
contractions that have also these electrical
165
00:09:33.520 --> 00:09:35.400
Emerson Keenan: patterns that exist in these signals.
166
00:09:35.400 --> 00:09:40.360
Emerson Keenan: So again feeding in that raw electrical data
and taking the outcomes that we know.
167
00:09:40.360 --> 00:09:42.720
Emerson Keenan: So if we follow this woman over time and work
out,
168
00:09:42.760 --> 00:09:46.440
Emerson Keenan: hey, the ones that went into delivery within
seven days had a certain kind of pattern.
169
00:09:46.480 --> 00:09:50.080
Emerson Keenan: Again, we can train the AI and better predict
these kind of outcomes.
170
00:09:50.080 --> 00:09:54.930
Emerson Keenan: And I think that's where AI is a really
transformative technology in that we can have
171
00:09:54.930 --> 00:09:57.610
Emerson Keenan: this expert level diagnosis and in some
cases,
172
00:09:57.610 --> 00:10:01.610
Emerson Keenan: even surpassing expert level diagnosis
available to anyone,
173
00:10:01.650 --> 00:10:03.410
Emerson Keenan: everyone, at any time.
174
00:10:03.410 --> 00:10:07.370
Emerson Keenan: And I think that's where it has these real
benefits, is having that really high accuracy
175
00:10:07.370 --> 00:10:09.090
Emerson Keenan: available at any time to the clinician.
176
00:10:09.410 --> 00:10:15.010
Sarath Ranganathan: In other bits of kit or a test that we use in
medicine we will refer to the positive
177
00:10:15.010 --> 00:10:17.770
Sarath Ranganathan: predictive value and negative predictive
value.
178
00:10:17.770 --> 00:10:20.090
Sarath Ranganathan: And look at receiver operating curves.
179
00:10:20.250 --> 00:10:25.890
Sarath Ranganathan: You're obviously using this device in
sometimes quite high risk situations or
180
00:10:25.890 --> 00:10:27.770
Sarath Ranganathan: situations that become high risk.
181
00:10:28.010 --> 00:10:31.170
Sarath Ranganathan: How confident are you in the performance of
the device?
182
00:10:31.530 --> 00:10:36.157
Sarath Ranganathan: Is it sort of comprehensive,
so it really does support the decision making
183
00:10:36.157 --> 00:10:40.810
Sarath Ranganathan: and I mean obstetrics is a fraught field so
can things go wrong?
184
00:10:41.330 --> 00:10:45.410
Emerson Keenan: Yeah, it's a fantastic question and a really
important thing we've looked at. So the
185
00:10:45.410 --> 00:10:49.660
Emerson Keenan: initial focus of technology providing that
data to the clinician means that they can
186
00:10:49.660 --> 00:10:53.060
Emerson Keenan: provide that clinical decision making
capacity in real time.
187
00:10:53.060 --> 00:10:56.900
Emerson Keenan: So the data is flowing in real time over the
internet to that clinician portal.
188
00:10:56.900 --> 00:11:00.260
Emerson Keenan: And they're looking at that data just like
they would if the patient was there in a
189
00:11:00.260 --> 00:11:04.220
Emerson Keenan: clinical setting. And so one of the real
valuable things we see is that sometimes we
190
00:11:04.260 --> 00:11:07.460
Emerson Keenan: find that women aren't able to attend these
appointments. So in these scenarios we don't
191
00:11:07.460 --> 00:11:10.940
Emerson Keenan: have the data at all. So if we can have this
ability to at least have that information
192
00:11:10.940 --> 00:11:13.300
Emerson Keenan: available, the clinician can then make that
decision.
193
00:11:13.300 --> 00:11:15.540
Emerson Keenan: And so if we do find something that's
concerning,
194
00:11:15.540 --> 00:11:17.740
Emerson Keenan: the doctor will say, hey,
we need you to come into hospital for further
195
00:11:17.740 --> 00:11:20.940
Emerson Keenan: monitoring or further checks. And so I think
that's a really important thing as part of
196
00:11:20.940 --> 00:11:24.260
Emerson Keenan: the clinical workflow. And then moving into
the predictive side of things,
197
00:11:24.540 --> 00:11:27.580
Emerson Keenan: the ability to have an AI,
predicting the outcomes,
198
00:11:27.580 --> 00:11:31.540
Emerson Keenan: we always want to be making sure that we are
better than what's currently available.
199
00:11:31.540 --> 00:11:34.940
Emerson Keenan: So, you know, there might be in the instance
of predicting the onset of labor.
200
00:11:34.940 --> 00:11:38.020
Emerson Keenan: Currently, there's a test called the fetal
fibronectin test,
201
00:11:38.420 --> 00:11:42.860
Emerson Keenan: which generally only has a positive
predictive value around 30 to 40%.
202
00:11:42.980 --> 00:11:47.580
Emerson Keenan: So if we can improve upon those statistics
through robust validation before this goes to
203
00:11:47.620 --> 00:11:49.990
Emerson Keenan: market showing we're greatly improving on
that.
204
00:11:50.070 --> 00:11:53.510
Emerson Keenan: That's really our goal to show that we can do
this better than the way things are doing
205
00:11:53.550 --> 00:11:54.190
Emerson Keenan: currently.
206
00:11:54.430 --> 00:11:59.070
Sarath Ranganathan: We're at the moment really striving to
connect people within this med school to
207
00:11:59.110 --> 00:12:04.630
Sarath Ranganathan: disciplines and professions and skill sets
outside of med school and other parts of
208
00:12:04.630 --> 00:12:08.230
Sarath Ranganathan: university, and more broadly in society
itself.
209
00:12:08.470 --> 00:12:13.910
Sarath Ranganathan: So I'm intrigued as to the process and the
conversations that enabled you to come into
210
00:12:13.910 --> 00:12:14.870
Sarath Ranganathan: our medical school.
211
00:12:14.910 --> 00:12:16.590
Emerson Keenan: Yeah, it's a really lovely story.
212
00:12:16.590 --> 00:12:21.070
Emerson Keenan: So very early in my PhD journey,
I said to my engineering supervisor,
213
00:12:21.110 --> 00:12:24.830
Emerson Keenan: you know, we need a clinician involved and
working through this with us.
214
00:12:24.870 --> 00:12:27.790
Emerson Keenan: You know, the only way we're going to
understand the clinical challenges and be
215
00:12:27.790 --> 00:12:31.190
Emerson Keenan: able to make this work is having a clinician
deeply embedded in the team.
216
00:12:31.190 --> 00:12:36.870
Emerson Keenan: And so he reached out to a doctor colleague
of his and was working in the neurology space
217
00:12:36.870 --> 00:12:40.030
Emerson Keenan: who, funnily enough, knew another obstetric
colleague,
218
00:12:40.190 --> 00:12:43.470
Emerson Keenan: who knew another obstetric colleague who had
the same idea as us,
219
00:12:43.470 --> 00:12:46.350
Emerson Keenan: which is now my co-founder and research
collaborator,
220
00:12:46.350 --> 00:12:47.600
Emerson Keenan: Doctor Fiona Brownfoot.
221
00:12:47.840 --> 00:12:50.880
Emerson Keenan: And so they kind of both said,
hey, you guys are thinking about a very
222
00:12:50.880 --> 00:12:54.560
Emerson Keenan: similar idea here. You should meet up and
talk further. And so we went out for a coffee
223
00:12:54.560 --> 00:12:58.440
Emerson Keenan: and started talking about the challenges with
with pregnancy monitoring and realized very
224
00:12:58.440 --> 00:13:01.920
Emerson Keenan: quickly we were on the same page. We both
wanted to solve the same problem just from
225
00:13:01.920 --> 00:13:05.920
Emerson Keenan: the opposite lens. We had an engineering
solution and they had that clinical need.
226
00:13:05.920 --> 00:13:09.520
Emerson Keenan: And I think it's such an important part of
developing technology that's actually
227
00:13:09.520 --> 00:13:13.400
Emerson Keenan: valuable for the clinical setting,
is having that clinical guidance really early
228
00:13:13.400 --> 00:13:18.680
Emerson Keenan: in the process. And I think the more we can
do as a school to enable that will be really
229
00:13:18.680 --> 00:13:21.760
Emerson Keenan: important. And I think from there,
it was interesting for me as an engineer,
230
00:13:21.800 --> 00:13:25.720
Emerson Keenan: working with Fiona throughout the course of
my PhD and just getting more and more
231
00:13:25.720 --> 00:13:29.800
Emerson Keenan: interested in that clinical application,
that I actually moved over to the medical
232
00:13:29.800 --> 00:13:33.280
Emerson Keenan: school. So I'm now an honorary research
fellow. I spent a bit of time as a postdoc in
233
00:13:33.320 --> 00:13:36.480
Emerson Keenan: the medical facility,
and it's for me, it's just so exciting to be
234
00:13:36.480 --> 00:13:39.280
Emerson Keenan: embedded in that clinical space. You know,
working in the clinic,
235
00:13:39.280 --> 00:13:42.360
Emerson Keenan: seeing how the technology is used. And it's
interesting as an engineer, often they're not
236
00:13:42.360 --> 00:13:44.920
Emerson Keenan: used how you think they are,
which is so important to see.
237
00:13:44.960 --> 00:13:47.680
Emerson Keenan: You might design it a certain way,
but in the reality of it,
238
00:13:47.680 --> 00:13:49.600
Emerson Keenan: things get used in the clinical workflows.
239
00:13:49.600 --> 00:13:53.000
Emerson Keenan: They need to be quick. They need to be easy
to understand. I think the more things we can
240
00:13:53.000 --> 00:13:57.280
Emerson Keenan: do to bring together engineers who have an
idea and clinicians have a problem,
241
00:13:57.680 --> 00:13:59.800
Emerson Keenan: it's going to be really great to see how that
all evolves.
242
00:13:59.960 --> 00:14:03.760
Sarath Ranganathan: I was going to ask you how you perceive
clinicians to think,
243
00:14:03.920 --> 00:14:07.040
Sarath Ranganathan: and you sort of summarized it there that
clinicians have just got problems.
244
00:14:07.080 --> 00:14:10.360
Emerson Keenan: Yeah, they've got problems and not enough
time. And you need to make it very easy to
245
00:14:10.400 --> 00:14:12.840
Emerson Keenan: solve the problem quickly and don't make it
hard.
246
00:14:13.040 --> 00:14:16.320
Emerson Keenan: If there's a 100 page instruction manual and
you've got to do 20 steps, it's just not
247
00:14:16.320 --> 00:14:19.000
Emerson Keenan: going to work. It's got to be quick,
easy and efficient to use.
248
00:14:19.040 --> 00:14:20.360
Sarath Ranganathan: Yeah, I would agree with that.
249
00:14:20.680 --> 00:14:24.360
Sarath Ranganathan: Now people often refer to the the black box
of algorithms,
250
00:14:24.360 --> 00:14:26.320
Sarath Ranganathan: particularly in predictive algorithms.
251
00:14:26.640 --> 00:14:31.040
Sarath Ranganathan: And there are commercial devices like on my
watch it can measure my heart rate.
252
00:14:31.320 --> 00:14:34.280
Sarath Ranganathan: Do you as an electrical engineer interested
in this area,
253
00:14:34.480 --> 00:14:40.760
Sarath Ranganathan: have some deep insights into how exactly it's
monitoring the heart rate there on my watch,
254
00:14:40.800 --> 00:14:46.970
Sarath Ranganathan: and therefore, are there any similarities in
that technology or the use case with you and
255
00:14:46.970 --> 00:14:49.010
Sarath Ranganathan: your device that can measure fetal heart
rate?
256
00:14:49.050 --> 00:14:52.410
Emerson Keenan: Yeah, it's absolutely fascinating,
the stuff that the smartwatch companies are
257
00:14:52.410 --> 00:14:56.290
Emerson Keenan: bringing out with these predictive analytics
and the data sets they've collected are
258
00:14:56.450 --> 00:15:00.690
Emerson Keenan: enormous. Like, I think I've seen an Apple
Watch study that was at least 50 or 60,000
259
00:15:00.690 --> 00:15:04.650
Emerson Keenan: participants looking at atrial fibrillation.
So there's a few elements to the
260
00:15:04.650 --> 00:15:07.170
Emerson Keenan: explainability. You know,
in the early days of AI,
261
00:15:07.570 --> 00:15:10.090
Emerson Keenan: I think we were just focusing on the
performance benefits. Just saying, hey, how
262
00:15:10.090 --> 00:15:13.770
Emerson Keenan: can we get the best performance possible? But
there has been a real shift to move to more
263
00:15:13.770 --> 00:15:16.770
Emerson Keenan: explainable AI techniques,
and there's been a lot of good research done
264
00:15:16.770 --> 00:15:19.890
Emerson Keenan: in how we can better understand what these
models are picking up on.
265
00:15:19.890 --> 00:15:24.450
Emerson Keenan: And so we've made methods that we can
actually project back onto the raw signals,
266
00:15:24.610 --> 00:15:26.770
Emerson Keenan: the things that the AI models are detecting.
267
00:15:27.370 --> 00:15:30.610
Emerson Keenan: So I'll talk briefly about our own use case
first and then move into,
268
00:15:30.650 --> 00:15:34.370
Emerson Keenan: I think, how the smartwatch might evolve in
the CTG analysis.
269
00:15:34.410 --> 00:15:38.850
Emerson Keenan: We actually published a study recently,
which we did this back propagation to show
270
00:15:38.890 --> 00:15:42.090
Emerson Keenan: what the AI was picking up,
and it was really fascinating without any
271
00:15:42.090 --> 00:15:47.340
Emerson Keenan: guidance or training. When we project the
features that are associated with fetal
272
00:15:47.340 --> 00:15:51.180
Emerson Keenan: distress. A lot of them were very similar
features to what clinicians associate with
273
00:15:51.180 --> 00:15:54.820
Emerson Keenan: fetal distress. So without any guidance on
what the physiology was or anything like
274
00:15:54.820 --> 00:15:59.100
Emerson Keenan: that, these highlights in the signals aligned
very closely with clinical guidance.
275
00:15:59.100 --> 00:16:02.300
Emerson Keenan: And so it was really interesting for us to
see that these models do pick up some of the
276
00:16:02.300 --> 00:16:05.220
Emerson Keenan: same things we've learned throughout years
and centuries,
277
00:16:05.220 --> 00:16:07.180
Emerson Keenan: in some cases of studying these signals.
278
00:16:07.180 --> 00:16:08.980
Emerson Keenan: So I think that's really interesting.
279
00:16:09.020 --> 00:16:12.540
Emerson Keenan: You know, seeing that reflected back to the
clinician when they go, hey, you know, some
280
00:16:12.540 --> 00:16:15.660
Emerson Keenan: of the things that's detecting are similar.
What I'm detecting is just highlighting them
281
00:16:15.660 --> 00:16:19.300
Emerson Keenan: for me and showing me really clearly what's
leading to that decision that's making. So I
282
00:16:19.300 --> 00:16:23.060
Emerson Keenan: think bringing explainable AI into these
methods is really important to get that
283
00:16:23.060 --> 00:16:24.980
Emerson Keenan: clinician confidence on what's going on.
284
00:16:25.380 --> 00:16:30.020
Emerson Keenan: In terms of the smartwatch side of things,
there's always a limitation around where
285
00:16:30.060 --> 00:16:32.540
Emerson Keenan: sensors are placed and what they can actually
pick up.
286
00:16:32.540 --> 00:16:37.420
Emerson Keenan: So I would love if we could have a pure
smartwatch that could measure everything we
287
00:16:37.420 --> 00:16:41.380
Emerson Keenan: need. But unfortunately the fetal signals are
very located to the abdomen.
288
00:16:41.380 --> 00:16:45.110
Emerson Keenan: So I think in this case they'll always need
to be some sort of abdominal based sensor
289
00:16:45.110 --> 00:16:47.070
Emerson Keenan: that measures that fetal component.
290
00:16:47.430 --> 00:16:50.950
Emerson Keenan: But I think the smartwatch does have a lot of
interesting applications looking at the
291
00:16:50.950 --> 00:16:53.030
Emerson Keenan: maternal side of things. Maternal heart rate.
292
00:16:53.390 --> 00:16:56.150
Emerson Keenan: There's been some early studies showing
association with maternal heart rate and
293
00:16:56.150 --> 00:17:00.030
Emerson Keenan: onset to delivery. So I think these platforms
are going to be really interesting as they
294
00:17:00.030 --> 00:17:03.030
Emerson Keenan: evolve. Having your smartwatch picking up so
much information,
295
00:17:03.030 --> 00:17:06.710
Emerson Keenan: and I think feeding that back to your doctor
and having that available is the next
296
00:17:06.710 --> 00:17:10.070
Emerson Keenan: frontier of clinical care. You know,
integrating these insights into all sorts of
297
00:17:10.070 --> 00:17:13.190
Emerson Keenan: medicine is somewhere,
I think, you know, both both I know the
298
00:17:13.190 --> 00:17:15.990
Emerson Keenan: companies themselves are looking into. But we
should think about more broadly how we
299
00:17:15.990 --> 00:17:17.630
Emerson Keenan: integrate this into medical practice.
300
00:17:17.670 --> 00:17:20.350
Sarath Ranganathan: Now I've got what potentially could be a dumb
question.
301
00:17:20.350 --> 00:17:23.630
Sarath Ranganathan: And if you think it's a dumb question,
you don't have to answer.
302
00:17:23.630 --> 00:17:28.430
Sarath Ranganathan: But I do recall a study many years ago was
done by a Swiss colleague of mine that looked
303
00:17:28.430 --> 00:17:32.630
Sarath Ranganathan: at exacerbations of asthma and showed that
they were predictable,
304
00:17:32.630 --> 00:17:34.310
Sarath Ranganathan: although in a very unusual way.
305
00:17:34.310 --> 00:17:37.550
Sarath Ranganathan: So the mathematics that predicted it was
fractals.
306
00:17:37.550 --> 00:17:39.110
Sarath Ranganathan: So sort of chaos theory.
307
00:17:39.110 --> 00:17:44.200
Sarath Ranganathan: And the question I have is and I've asked
this of other AI experts is are the
308
00:17:44.200 --> 00:17:50.160
Sarath Ranganathan: algorithms when you actually analyze them,
are they sort of fractal type mathematics or
309
00:17:50.160 --> 00:17:53.480
Sarath Ranganathan: does anyone know or is it really genuinely a
black box?
310
00:17:54.560 --> 00:17:56.680
Emerson Keenan: Oh, it's a it's a tough question.
311
00:17:57.440 --> 00:18:00.080
Emerson Keenan: It depends on the type of model you're
training.
312
00:18:00.080 --> 00:18:04.600
Emerson Keenan: So different model structures will pick up on
different kinds of information at a very high
313
00:18:04.600 --> 00:18:08.440
Emerson Keenan: level. You know, you might have things like a
convolutional network with a certain window
314
00:18:08.440 --> 00:18:10.680
Emerson Keenan: size, and they will only look at a very,
you know,
315
00:18:10.720 --> 00:18:12.600
Emerson Keenan: restricted piece of the signal.
316
00:18:12.600 --> 00:18:16.600
Emerson Keenan: Whereas some other model architectures like
your LSTMs or a transformer model,
317
00:18:16.600 --> 00:18:19.520
Emerson Keenan: do look at these long range patterns and
dependencies.
318
00:18:19.520 --> 00:18:23.480
Emerson Keenan: So that might be somewhere where the model
structure itself can determine what you're
319
00:18:23.520 --> 00:18:25.080
Emerson Keenan: actually picking up in the signal.
320
00:18:25.080 --> 00:18:27.600
Emerson Keenan: So I think there is always a bit of,
you know,
321
00:18:27.640 --> 00:18:31.400
Emerson Keenan: human nous, or at least currently to know
what the best kind of model structure is for
322
00:18:31.400 --> 00:18:34.800
Emerson Keenan: a given task and knowing the structure of the
physiological signal.
323
00:18:34.800 --> 00:18:38.480
Emerson Keenan: So I think it's interesting. I think these
things knowing the fractal dynamic can guide
324
00:18:38.480 --> 00:18:42.530
Emerson Keenan: you then to make the right AI model decision
to get the best result.
325
00:18:42.930 --> 00:18:44.090
Sarath Ranganathan: Emerson, you're so polite.
326
00:18:44.130 --> 00:18:47.170
Sarath Ranganathan: You said that was a tough question. I don't
know if that's another way,
327
00:18:47.210 --> 00:18:49.450
Sarath Ranganathan: politely, of saying it was actually a dumb
question,
328
00:18:49.450 --> 00:18:57.130
Sarath Ranganathan: but. But I've come across you before and in
that meeting where I met you,
329
00:18:57.130 --> 00:18:59.210
Sarath Ranganathan: you were giving a wonderful presentation.
330
00:18:59.210 --> 00:19:02.570
Sarath Ranganathan: You had explored AI technologies for
research.
331
00:19:02.570 --> 00:19:06.370
Sarath Ranganathan: And I think you've also done a little bit in
terms of other aspects of healthcare related
332
00:19:06.370 --> 00:19:11.155
Sarath Ranganathan: to Kali, etc.. How do you see,
because I see you as an expert here,
333
00:19:11.155 --> 00:19:15.570
Sarath Ranganathan: how do you see AI technologies continuing to
evolve in healthcare?
334
00:19:15.570 --> 00:19:17.650
Sarath Ranganathan: Can you predict what's going to happen?
335
00:19:18.130 --> 00:19:21.130
Emerson Keenan: I definitely can't predict it,
but I think there's some there's some early
336
00:19:21.130 --> 00:19:22.850
Emerson Keenan: signs about where things might go.
337
00:19:23.050 --> 00:19:25.970
Emerson Keenan: And I'm really excited by the potentials that
all of this is bringing.
338
00:19:26.290 --> 00:19:30.170
Emerson Keenan: You know, some of the techniques we've been
working on have been around for a while.
339
00:19:30.170 --> 00:19:33.810
Emerson Keenan: This ability to do the taking in input data
and predicting what,
340
00:19:33.810 --> 00:19:38.130
Emerson Keenan: what might come next. But the emergence of
generative AI techniques and large language
341
00:19:38.130 --> 00:19:42.740
Emerson Keenan: models has just really unlocked a huge area
of exploration.
342
00:19:42.740 --> 00:19:46.100
Emerson Keenan: I think one of the things that's most
exciting to me is the area of scientific
343
00:19:46.100 --> 00:19:49.820
Emerson Keenan: discovery. You know, the ability of these
tools to bring in large,
344
00:19:49.820 --> 00:19:53.980
Emerson Keenan: large amounts of information and really
concisely pick out the things that might be
345
00:19:53.980 --> 00:19:56.980
Emerson Keenan: really important. And I think one of the
things I've mentioned in a few of my
346
00:19:56.980 --> 00:20:01.900
Emerson Keenan: presentations is Google recently developed
what they call an AI code scientist,
347
00:20:02.060 --> 00:20:06.940
Emerson Keenan: and the idea of this model was that we can
give it some indication of an area we might
348
00:20:06.940 --> 00:20:10.380
Emerson Keenan: want to look into further,
and it goes away and generates hypotheses,
349
00:20:10.420 --> 00:20:13.940
Emerson Keenan: compares those hypotheses against data that's
already in the literature, and kind of does
350
00:20:13.940 --> 00:20:18.340
Emerson Keenan: this continual ranking to kind of narrow down
what might be the most attractive hypothesis
351
00:20:18.340 --> 00:20:20.140
Emerson Keenan: based on everything it's looked at.
352
00:20:20.140 --> 00:20:23.700
Emerson Keenan: And some of those early studies I know they
were looking at trying to find a drug
353
00:20:23.700 --> 00:20:28.980
Emerson Keenan: repurposing for a novel tumor candidate,
and they ran the the AI code scientists,
354
00:20:28.980 --> 00:20:30.660
Emerson Keenan: and it came up with a candidate.
355
00:20:30.660 --> 00:20:33.780
Emerson Keenan: And the scientists then went and validated
that there was nothing in the literature that
356
00:20:33.780 --> 00:20:37.260
Emerson Keenan: had ever used this drug candidate for that
purpose before, but then it validated,
357
00:20:37.300 --> 00:20:39.670
Emerson Keenan: they found that it could treat that tumor.
So, you know,
358
00:20:39.710 --> 00:20:42.830
Emerson Keenan: these kind of things where we,
you know, set the AI in a certain direction.
359
00:20:42.830 --> 00:20:45.550
Emerson Keenan: We say, hey, we've got this really
interesting problem space. I want you to go
360
00:20:45.550 --> 00:20:49.470
Emerson Keenan: and look at everything that's out there and
tell me what's the most promising candidate.
361
00:20:49.510 --> 00:20:50.830
Emerson Keenan: I'm just I'm just so excited.
362
00:20:50.830 --> 00:20:54.230
Emerson Keenan: The ability that we could narrow down these
things really quickly to know where we need
363
00:20:54.230 --> 00:20:57.150
Emerson Keenan: to go. And I think that's just such an
exciting area to move into.
364
00:20:57.430 --> 00:20:59.550
Sarath Ranganathan: With that positive example in mind.
365
00:20:59.590 --> 00:21:04.070
Sarath Ranganathan: What advice would you give to early career
researchers who are working or anticipating
366
00:21:04.070 --> 00:21:05.870
Sarath Ranganathan: working in AI or digital health?
367
00:21:05.910 --> 00:21:11.470
Emerson Keenan: The main advice I give when when I hear this
question is I think AI is coming.
368
00:21:11.470 --> 00:21:15.190
Emerson Keenan: It's much like the internet,
smartphones, these tools.
369
00:21:15.390 --> 00:21:18.310
Emerson Keenan: You know, there was this early days when a
few people were kind of using them and
370
00:21:18.310 --> 00:21:21.550
Emerson Keenan: adopting them, and all of a sudden we look
around the world, they're everywhere. They're
371
00:21:21.590 --> 00:21:22.990
Emerson Keenan: a necessity. Part of doing work.
372
00:21:22.990 --> 00:21:24.790
Emerson Keenan: You can't do work without using them.
373
00:21:24.790 --> 00:21:28.190
Emerson Keenan: So I think we need to treat it in the same
way that they are going to become part of the
374
00:21:28.190 --> 00:21:31.150
Emerson Keenan: way we do work, and we just need to start
using them,
375
00:21:31.150 --> 00:21:35.070
Emerson Keenan: start learning them, adopting our practices,
because the ones who do adopt them are going
376
00:21:35.070 --> 00:21:37.230
Emerson Keenan: to have an unfair advantage on those that
don't.
377
00:21:37.550 --> 00:21:41.750
Emerson Keenan: And the sooner you can get into the practice
of using them and working out how they work
378
00:21:41.750 --> 00:21:43.430
Emerson Keenan: will make your practice a lot easier.
379
00:21:43.430 --> 00:21:46.710
Emerson Keenan: And I think some of the the guiding things I
would say in doing this are,
380
00:21:46.990 --> 00:21:49.590
Emerson Keenan: you know, when when you go to use a new AI
technique,
381
00:21:49.710 --> 00:21:52.590
Emerson Keenan: think very carefully about the problem you
want to apply it for.
382
00:21:52.630 --> 00:21:57.150
Emerson Keenan: First, think about all the problems you might
use it on and find one that really is
383
00:21:57.150 --> 00:21:59.110
Emerson Keenan: amenable to being solved using AI.
384
00:21:59.150 --> 00:22:01.870
Emerson Keenan: Don't just have AI and go,
hey, I'm going to put it on every single
385
00:22:01.870 --> 00:22:05.030
Emerson Keenan: problem. I know it's finding where it's most
suited to that problem.
386
00:22:05.030 --> 00:22:08.390
Emerson Keenan: So I think really defining that really
clearly upfront is really important.
387
00:22:08.390 --> 00:22:11.390
Emerson Keenan: And I think things like the community of
practice. I know there's a community of
388
00:22:11.390 --> 00:22:14.710
Emerson Keenan: practice here at the university looking at
how we can use these AI techniques, you know,
389
00:22:14.750 --> 00:22:18.830
Emerson Keenan: developing these things in your lab groups,
in your your friendships and collegial
390
00:22:18.830 --> 00:22:22.270
Emerson Keenan: networks on the best practice of using these
things. And the more you can do that,
391
00:22:22.310 --> 00:22:26.630
Emerson Keenan: the faster ahead of the curve you will be as
they become mainstay until you know they are
392
00:22:26.670 --> 00:22:28.310
Emerson Keenan: just a way we do our work.
393
00:22:28.590 --> 00:22:31.550
Sarath Ranganathan: So I've got another tough replica question.
394
00:22:31.950 --> 00:22:35.910
Sarath Ranganathan: What advice would you give to the head of the
medical school about how we should be
395
00:22:36.080 --> 00:22:39.320
Sarath Ranganathan: approaching AI or accommodating developments
in AI?
396
00:22:39.840 --> 00:22:41.640
Emerson Keenan: So it's an interesting one,
I think.
397
00:22:41.680 --> 00:22:45.920
Emerson Keenan: I think encouraging this practice and making
it easier for people to access these tools is
398
00:22:45.920 --> 00:22:49.240
Emerson Keenan: really important. I know only the university
has an internal platform, Spark,
399
00:22:49.440 --> 00:22:51.080
Emerson Keenan: where researchers can can use this.
400
00:22:51.080 --> 00:22:54.200
Emerson Keenan: So I think really prioritizing access to
these tools,
401
00:22:54.520 --> 00:22:56.600
Emerson Keenan: training on these tools is really important.
402
00:22:56.760 --> 00:23:00.280
Emerson Keenan: I'm not sure if this is 100% the way to go,
but I know some of the big companies even
403
00:23:00.280 --> 00:23:03.040
Emerson Keenan: have leaderboards on who is using these tools
the most,
404
00:23:03.040 --> 00:23:06.320
Emerson Keenan: and having bonuses and prizes for who's
adopting these tools the most.
405
00:23:06.320 --> 00:23:10.120
Emerson Keenan: So that's an interesting way of doing things.
You know, that way you're incentivizing the
406
00:23:10.120 --> 00:23:13.360
Emerson Keenan: use of these tools. I think it could be
something really interesting to explore.
407
00:23:13.400 --> 00:23:15.520
Sarath Ranganathan: Well, I can certainly afford a virtual prize,
408
00:23:15.920 --> 00:23:18.040
Sarath Ranganathan: so I use AI in Spotify.
409
00:23:18.200 --> 00:23:20.120
Sarath Ranganathan: It chooses some of the songs I'd like.
410
00:23:20.160 --> 00:23:21.600
Sarath Ranganathan: I presume that's that's AI.
411
00:23:21.640 --> 00:23:26.040
Sarath Ranganathan: So I'm going to move on. Now to the final
question is what are you listening to right
412
00:23:26.040 --> 00:23:31.520
Sarath Ranganathan: now, Emerson? And how does it relate to your
work and your life experiences?
413
00:23:31.560 --> 00:23:32.680
Sarath Ranganathan: That's a complex question.
414
00:23:32.720 --> 00:23:37.675
Emerson Keenan: Interesting. So actually I'm listening to the
newest album by a band called Geese lately,
415
00:23:37.675 --> 00:23:42.610
Emerson Keenan: like the bird. And the song I really like off
that album is a song called Husbands.
416
00:23:43.250 --> 00:23:45.810
Emerson Keenan: It's difficult to explain without listening
to the track itself, but it's super
417
00:23:45.810 --> 00:23:48.330
Emerson Keenan: interesting melodically and rhythmically.
418
00:23:48.330 --> 00:23:51.850
Emerson Keenan: So there's, you know,
all these different guitar parts happening.
419
00:23:51.890 --> 00:23:55.770
Emerson Keenan: There's a very interesting drum line,
and it's all kind of weaved together in this
420
00:23:55.770 --> 00:23:59.570
Emerson Keenan: really complex way. And I think,
I think it's really beautiful when,
421
00:23:59.570 --> 00:24:03.250
Emerson Keenan: when all these things can come together in a
really cohesive way.
422
00:24:03.250 --> 00:24:05.970
Emerson Keenan: And I think bringing together,
you know, different skill sets,
423
00:24:06.010 --> 00:24:08.730
Emerson Keenan: engineering, science,
medicine, the human element,
424
00:24:08.730 --> 00:24:12.530
Emerson Keenan: I think is really resonant of what we should
be trying to do in the medical school and in
425
00:24:12.530 --> 00:24:16.530
Emerson Keenan: biomedical research more generally is how do
we bring all these tools together?
426
00:24:16.570 --> 00:24:20.690
Emerson Keenan: Not forgetting the human element,
you know, thinking about how it makes us do
427
00:24:20.690 --> 00:24:23.770
Emerson Keenan: our jobs better, live our lives better,
and you know,
428
00:24:23.810 --> 00:24:24.890
Emerson Keenan: develop a better world.
429
00:24:25.290 --> 00:24:27.130
Sarath Ranganathan: Well, it's been lovely talking to you.
430
00:24:27.170 --> 00:24:29.130
Sarath Ranganathan: Congratulations on Kali Healthcare.
431
00:24:29.130 --> 00:24:31.650
Sarath Ranganathan: I hope that you have every success with it.
432
00:24:31.770 --> 00:24:37.300
Sarath Ranganathan: I think you're a great example of you know,
involving someone from outside the medical or
433
00:24:37.300 --> 00:24:38.820
Sarath Ranganathan: health disciplines in health.
434
00:24:38.820 --> 00:24:41.700
Sarath Ranganathan: And just to see what can be achieved through
that,
435
00:24:41.700 --> 00:24:45.900
Sarath Ranganathan: I think can have a wonderful future in our
school. I hope you stay within our school and
436
00:24:45.940 --> 00:24:49.500
Sarath Ranganathan: that we build a community of people with a
similar interests as you.
437
00:24:49.500 --> 00:24:53.380
Sarath Ranganathan: So thanks for your time and I look forward to
listening to Geese.
438
00:24:53.740 --> 00:24:55.460
Emerson Keenan: Thanks. It's been a pleasure being on the
show.
439
00:24:57.020 --> 00:24:59.300
Sarath Ranganathan: Thanks for joining us on today's episode.
440
00:24:59.900 --> 00:25:03.020
Sarath Ranganathan: If you have a question or a brilliant idea
for a future topic,
441
00:25:03.020 --> 00:25:06.140
Sarath Ranganathan: we'd love to hear from you via the link in
the show notes.
442
00:25:06.860 --> 00:25:10.220
Sarath Ranganathan: And feel free to share this episode with a
friend or colleague,
443
00:25:10.380 --> 00:25:11.980
Sarath Ranganathan: or even a member of your family.
444
00:25:12.460 --> 00:25:13.500
Sarath Ranganathan: See you next time!
1
00:00:04.120 --> 00:00:07.240
Sarath Ranganathan: Whether you're a student,
staff member, alumni,
2
00:00:07.240 --> 00:00:10.520
Sarath Ranganathan: or someone who's passionate about healthcare
and its future,
3
00:00:10.920 --> 00:00:13.400
Sarath Ranganathan: you're listening to the right podcast.
4
00:00:21.560 --> 00:00:26.320
Sarath Ranganathan: This podcast was recorded on the lands of the
Wurundjeri people of the Kulin nation,
5
00:00:26.680 --> 00:00:30.760
Sarath Ranganathan: and we pay our respects to elders,
past and present.
6
00:00:36.320 --> 00:00:38.215
Sarath Ranganathan: Welcome to MMS Network.
7
00:00:38.215 --> 00:00:41.956
Sarath Ranganathan: I'm your host, Professor Sarah Ranganathan,
head of the Melbourne Medical School,
8
00:00:41.956 --> 00:00:46.680
Sarath Ranganathan: and today we'll be discussing one of the most
transformative forces in modern medicine,
9
00:00:47.160 --> 00:00:49.360
Sarath Ranganathan: that is artificial intelligence.
10
00:00:49.840 --> 00:00:52.720
Sarath Ranganathan: And joining me in the studio is Doctor
Emerson Keenan,
11
00:00:52.880 --> 00:00:58.240
Sarath Ranganathan: co-founder of Kali Healthcare and a research
fellow in our department of Obstetrics,
12
00:00:58.400 --> 00:01:00.690
Sarath Ranganathan: Gynecology and Newborn Health.
13
00:01:01.090 --> 00:01:05.010
Sarath Ranganathan: Emerson's work sits at the intersection of AI
and medical devices,
14
00:01:05.370 --> 00:01:09.530
Sarath Ranganathan: reshaping how we think about patient care and
clinical research.
15
00:01:10.410 --> 00:01:12.650
Sarath Ranganathan: Welcome, Emerson. It's great to have you
here.
16
00:01:13.370 --> 00:01:17.490
Sarath Ranganathan: To start us off, could you tell us a little
bit more about your background?
17
00:01:17.690 --> 00:01:21.210
Emerson Keenan: Fantastic. Thanks so much for having me,
Sarath. So I'm actually an electrical
18
00:01:21.210 --> 00:01:22.330
Emerson Keenan: engineer by background.
19
00:01:22.330 --> 00:01:27.010
Emerson Keenan: So I did my original undergraduate studies in
Brisbane in electrical engineering,
20
00:01:27.010 --> 00:01:31.290
Emerson Keenan: and my first taste of medicine was in my
honours year doing a biomedical project,
21
00:01:31.290 --> 00:01:34.770
Emerson Keenan: looking at a smartphone application for
detecting sleep apnea.
22
00:01:34.770 --> 00:01:36.250
Emerson Keenan: And so that was really exciting.
23
00:01:36.250 --> 00:01:39.850
Emerson Keenan: And they went on to be commercialized through
a company called ResApp. And so from there
24
00:01:39.890 --> 00:01:41.690
Emerson Keenan: kind of sparked my interest in medicine.
25
00:01:41.690 --> 00:01:46.370
Emerson Keenan: So from there, I began a PhD at the
University of Melbourne looking at new
26
00:01:46.370 --> 00:01:50.530
Emerson Keenan: technologies for pregnancy monitoring. And so
that evolved over time to now spin out into
27
00:01:50.570 --> 00:01:54.490
Emerson Keenan: Kali Healthcare. And so I also have an
ongoing appointment as an honorary research
28
00:01:54.490 --> 00:01:57.290
Emerson Keenan: fellow in the Melbourne Medical School,
looking at how we apply artificial
29
00:01:57.290 --> 00:02:00.220
Emerson Keenan: intelligence to things like wearables,
large data sets,
30
00:02:00.260 --> 00:02:02.580
Emerson Keenan: biomedical data, and it's a really exciting
space to be in.
31
00:02:02.820 --> 00:02:07.260
Sarath Ranganathan: So when you started electrical engineering,
did you have any idea then that this was the
32
00:02:07.260 --> 00:02:09.180
Sarath Ranganathan: journey you were going to find yourself on?
33
00:02:09.220 --> 00:02:13.020
Emerson Keenan: No, I think I think back then I was just
really excited by what technology can do and
34
00:02:13.020 --> 00:02:17.180
Emerson Keenan: the power of humans to use technology to
solve really important problems.
35
00:02:17.180 --> 00:02:19.540
Emerson Keenan: But over time, I found more and more that,
you know,
36
00:02:19.540 --> 00:02:21.700
Emerson Keenan: solving these problems is one of the most
impactful things I could do with that
37
00:02:21.700 --> 00:02:22.940
Emerson Keenan: electrical engineering skill.
38
00:02:23.140 --> 00:02:27.060
Sarath Ranganathan: So tell us a little bit more about what your
PhD focused on,
39
00:02:27.260 --> 00:02:30.220
Sarath Ranganathan: and also how it evolved into Kali Healthcare.
40
00:02:30.580 --> 00:02:35.020
Emerson Keenan: My PhD was originally looking at ways that we
can better monitor pregnancies,
41
00:02:35.060 --> 00:02:36.620
Emerson Keenan: ideally using wearable devices.
42
00:02:36.620 --> 00:02:39.060
Emerson Keenan: So the crux of it is that the current
technology,
43
00:02:39.060 --> 00:02:41.940
Emerson Keenan: which uses ultrasound,
is a really hospital centric approach.
44
00:02:42.100 --> 00:02:45.700
Emerson Keenan: So women who have a pregnancy complication
are coming into hospitals regularly to have
45
00:02:45.700 --> 00:02:48.100
Emerson Keenan: monitoring done. It's really time consuming
for women,
46
00:02:48.100 --> 00:02:51.540
Emerson Keenan: time consuming for clinicians. So we looked
at ways of technologies that could be used
47
00:02:51.540 --> 00:02:54.580
Emerson Keenan: outside of that clinical setting so that we
can free up clinician time.
48
00:02:54.780 --> 00:02:58.470
Emerson Keenan: And so throughout my PhD,
I found that electrical signals seem to be a
49
00:02:58.470 --> 00:03:02.230
Emerson Keenan: really promising way of doing this monitoring
without needing a clinician to be physically
50
00:03:02.230 --> 00:03:05.790
Emerson Keenan: present. So originally it was looking at
developing the sensor patch,
51
00:03:05.790 --> 00:03:08.710
Emerson Keenan: developing the layout and the algorithms to
process things.
52
00:03:08.710 --> 00:03:12.230
Emerson Keenan: And over time, we found that by putting these
sensors in the right place and the right
53
00:03:12.230 --> 00:03:16.270
Emerson Keenan: configuration with the right algorithms,
which eventually become an artificial
54
00:03:16.270 --> 00:03:19.230
Emerson Keenan: intelligence algorithm,
we could get really reliable results and
55
00:03:19.230 --> 00:03:21.390
Emerson Keenan: unlock this ability to do monitoring and home
safely.
56
00:03:21.710 --> 00:03:26.270
Sarath Ranganathan: So how does the device work and how do you
integrate that into clinical care for
57
00:03:26.310 --> 00:03:27.230
Sarath Ranganathan: monitoring women?
58
00:03:27.350 --> 00:03:32.350
Emerson Keenan: Yeah. So the device essentially is a smart
device which uses a wearable sensor patch.
59
00:03:32.470 --> 00:03:35.510
Emerson Keenan: So we have that wearable sensor that goes on
the mum's abdomen.
60
00:03:35.510 --> 00:03:39.590
Emerson Keenan: There's a small unit attached to it which has
all the data collection infrastructure.
61
00:03:39.590 --> 00:03:41.190
Emerson Keenan: So we can pick up those signals.
62
00:03:41.190 --> 00:03:44.670
Emerson Keenan: It then uses Bluetooth to send the
information to the mum's smartphone.
63
00:03:44.710 --> 00:03:48.270
Emerson Keenan: That smartphone helps guide them through the
process of putting on the device,
64
00:03:48.310 --> 00:03:51.470
Emerson Keenan: getting that measurement reading,
as well as then sending that information over
65
00:03:51.470 --> 00:03:55.350
Emerson Keenan: the internet back to the clinician where they
can do a review. So it's really that
66
00:03:55.350 --> 00:03:58.880
Emerson Keenan: combination of that, that wearable device,
a smartphone that helps guide the user
67
00:03:58.920 --> 00:04:02.600
Emerson Keenan: through the process, as well as that internet
enabled platform that makes this all work.
68
00:04:03.040 --> 00:04:08.840
Sarath Ranganathan: So what sort of decisions do the clinicians
make based on what data or advice the device
69
00:04:08.880 --> 00:04:09.920
Sarath Ranganathan: is sending to them?
70
00:04:10.240 --> 00:04:14.840
Emerson Keenan: The initial focus of our technology is to
unlock new cases where we can use this.
71
00:04:14.840 --> 00:04:17.960
Emerson Keenan: So having that remote monitoring is really
our first launch point.
72
00:04:17.960 --> 00:04:22.320
Emerson Keenan: So the ability to unlock this care at home
freeing up clinician time's really valuable.
73
00:04:22.320 --> 00:04:26.040
Emerson Keenan: So the idea is we're giving them the same
information they're used to seeing from that
74
00:04:26.040 --> 00:04:29.440
Emerson Keenan: clinical setting. So that's in the initial
instance the fetal heart rate,
75
00:04:29.480 --> 00:04:32.320
Emerson Keenan: the maternal heart rate and the measure of
the contractions.
76
00:04:32.320 --> 00:04:34.640
Emerson Keenan: And they can see that all in the patient
dashboard.
77
00:04:34.760 --> 00:04:38.680
Emerson Keenan: We're working over time to bring more things
into the platform. So looking at bringing in
78
00:04:38.680 --> 00:04:41.960
Emerson Keenan: things like continuous glucose monitoring,
blood pressure monitoring,
79
00:04:42.000 --> 00:04:43.920
Emerson Keenan: having this all part of that platform.
80
00:04:43.920 --> 00:04:48.760
Emerson Keenan: And then really the grand vision behind all
of this is to add clinical decision support
81
00:04:48.760 --> 00:04:52.000
Emerson Keenan: into it. So not only providing that data over
the internet,
82
00:04:52.000 --> 00:04:55.280
Emerson Keenan: but having the ability to predict things
before they eventuate.
83
00:04:55.280 --> 00:05:00.120
Emerson Keenan: So one of the first things we're looking at
is using the uterine contraction information
84
00:05:00.120 --> 00:05:04.000
Emerson Keenan: to better pick up when women are going to go
into labor and predicting if some people
85
00:05:04.000 --> 00:05:07.120
Emerson Keenan: have, particularly in early labor and maybe
have preterm birth.
86
00:05:07.360 --> 00:05:10.600
Emerson Keenan: So we've got some early evidence that shows
using AI techniques,
87
00:05:10.600 --> 00:05:13.600
Emerson Keenan: that we can predict these cases where women
are going into labor early,
88
00:05:13.640 --> 00:05:15.320
Emerson Keenan: up to seven days in advance.
89
00:05:15.320 --> 00:05:17.760
Emerson Keenan: So having something like that,
if a doctor's doing a remote monitoring
90
00:05:17.760 --> 00:05:20.160
Emerson Keenan: session for a complicated pregnancy and they
say,
91
00:05:20.160 --> 00:05:23.440
Emerson Keenan: hey, there's a really strong likelihood that
you're going to go into delivery, they can
92
00:05:23.440 --> 00:05:26.360
Emerson Keenan: better triage that care. You know,
they can bring them into hospital, monitor
93
00:05:26.360 --> 00:05:30.760
Emerson Keenan: them more closely, and really only bring the
women in who are at the highest risk into the
94
00:05:30.760 --> 00:05:34.320
Emerson Keenan: hospital setting, and everyone else can stay
at home safely. So we're really excited about
95
00:05:34.600 --> 00:05:37.680
Emerson Keenan: where this can go beyond just unlocking those
new models of care,
96
00:05:37.720 --> 00:05:39.640
Emerson Keenan: moving into that predictive analytics space.
97
00:05:40.160 --> 00:05:43.880
Sarath Ranganathan: So from the mother's point of view,
someone who's wearing the device,
98
00:05:43.880 --> 00:05:49.560
Sarath Ranganathan: what do they notice? Or what would they tell
you about the benefits of that device?
99
00:05:49.600 --> 00:05:53.560
Emerson Keenan: Yeah. So throughout my PhD,
I assisted with a lot of the data collection.
100
00:05:53.560 --> 00:05:56.770
Emerson Keenan: So I was there in the clinic,
helping set up the device,
101
00:05:56.770 --> 00:06:00.370
Emerson Keenan: talking to women about their experiences. And
it was really incredible to see,
102
00:06:00.410 --> 00:06:04.690
Emerson Keenan: you know, just the reoccurring theme of how
much impact it has on women's lives having to
103
00:06:04.730 --> 00:06:08.690
Emerson Keenan: come in for this monitoring repeatedly,
the anxiety it brings when they're unsure at
104
00:06:08.690 --> 00:06:11.690
Emerson Keenan: home, they're thinking,
oh, should I come in, should I stay? And I
105
00:06:11.690 --> 00:06:15.850
Emerson Keenan: think that's a really important thing that we
think we can solve with this technology.
106
00:06:15.850 --> 00:06:19.610
Emerson Keenan: And so we actually incorporated this into one
of our most recent research studies,
107
00:06:19.610 --> 00:06:23.210
Emerson Keenan: where we asked women about their experiences
in having fetal monitoring.
108
00:06:23.210 --> 00:06:26.690
Emerson Keenan: And we published this recently in Nature
Digital Medicine,
109
00:06:27.090 --> 00:06:29.850
Emerson Keenan: which found that over 90% of women were
saying,
110
00:06:29.890 --> 00:06:33.610
Emerson Keenan: you know, the ability of this to reduce that
anxiety and the desire for having remote
111
00:06:33.610 --> 00:06:35.090
Emerson Keenan: monitoring was really high.
112
00:06:35.410 --> 00:06:38.570
Emerson Keenan: And so we think that,
you know, that that lens from the user
113
00:06:38.570 --> 00:06:41.130
Emerson Keenan: perspective is really important. It's not
just, you know,
114
00:06:41.170 --> 00:06:44.690
Emerson Keenan: solving some of these clinical workflow
challenges. It's also solving for mums that
115
00:06:44.690 --> 00:06:48.330
Emerson Keenan: that anxiety that, you know,
burden on them having to come into the
116
00:06:48.330 --> 00:06:50.930
Emerson Keenan: clinic, and particularly women who've got
kids at home,
117
00:06:50.970 --> 00:06:54.260
Emerson Keenan: you know, finding childcare for the kids or
finding people who can take care of other
118
00:06:54.260 --> 00:06:56.260
Emerson Keenan: family members. You know,
freeing all that up.
119
00:06:56.260 --> 00:06:59.740
Emerson Keenan: I think you know that the human aspect of the
technology is really important, you know,
120
00:06:59.780 --> 00:07:03.980
Emerson Keenan: helping solve not only clinical challenges,
but human usability challenges as well.
121
00:07:04.420 --> 00:07:06.540
Sarath Ranganathan: Now, that sounds incredibly positive.
122
00:07:07.140 --> 00:07:10.740
Sarath Ranganathan: Do you mind me then asking why you called
this Kali Healthcare,
123
00:07:10.780 --> 00:07:13.220
Sarath Ranganathan: given what I know about the goddess Kali?
124
00:07:13.260 --> 00:07:16.940
Emerson Keenan: Yeah. So originally the company was founded
by myself,
125
00:07:17.060 --> 00:07:19.940
Emerson Keenan: my engineering supervisor,
Marimuthu Palaniswami,
126
00:07:20.140 --> 00:07:22.660
Emerson Keenan: and my clinical colleague,
Doctor Fiona Brownfoot,
127
00:07:22.700 --> 00:07:24.780
Emerson Keenan: who's also in the obstetrics department.
128
00:07:24.780 --> 00:07:30.980
Emerson Keenan: And so when we're talking about ideas behind
the naming of the company Professor Marimuthu
129
00:07:30.980 --> 00:07:32.860
Emerson Keenan: Palaniswami suggested the goddess Kali.
130
00:07:32.860 --> 00:07:36.620
Emerson Keenan: And in his region, it's seen as a goddess of
fertility and motherhood.
131
00:07:36.620 --> 00:07:39.420
Emerson Keenan: And so we thought there was a really fitting
thing to,
132
00:07:39.460 --> 00:07:42.500
Emerson Keenan: you know, put behind the technology and
having that protective element.
133
00:07:42.860 --> 00:07:47.020
Sarath Ranganathan: Thanks for the explanation. I obviously
didn't understand the full skill sets of the
134
00:07:47.060 --> 00:07:51.380
Sarath Ranganathan: goddess Kali. So it's not really related to
AI itself,
135
00:07:51.380 --> 00:07:56.150
Sarath Ranganathan: but I did want to come on to ask about what
specific role AI now plays in that
136
00:07:56.150 --> 00:07:59.230
Sarath Ranganathan: monitoring. How did you integrate that into
the tool that you have?
137
00:07:59.470 --> 00:08:01.870
Emerson Keenan: So there's two key aspects of how we're using
AI.
138
00:08:01.910 --> 00:08:05.870
Emerson Keenan: So the first is the underlying processing of
the sensor data we collect.
139
00:08:05.870 --> 00:08:09.190
Emerson Keenan: So the sensor patch collects electrical
signals.
140
00:08:09.190 --> 00:08:13.590
Emerson Keenan: So all it's measuring is how those electrical
signals rise up and down on the abdomen.
141
00:08:13.590 --> 00:08:16.350
Emerson Keenan: And so that signal contains a really large
mixture.
142
00:08:16.350 --> 00:08:19.230
Emerson Keenan: It's got everything from the mum's heart rate
activity,
143
00:08:19.270 --> 00:08:22.630
Emerson Keenan: the baby's heart rate,
the muscle contractions of the uterus,
144
00:08:23.110 --> 00:08:26.630
Emerson Keenan: the other muscle contractions as they might
be moving around. And so it's really
145
00:08:26.630 --> 00:08:29.870
Emerson Keenan: difficult to separate all this out and unlock
those different signals.
146
00:08:29.870 --> 00:08:34.070
Emerson Keenan: And so we tried early in my PhD kind of
traditional approaches to extract this
147
00:08:34.070 --> 00:08:37.550
Emerson Keenan: information. And we just couldn't quite get
the reliability that we needed. And so
148
00:08:37.870 --> 00:08:40.950
Emerson Keenan: looking at AI gave us that breakthrough we
were looking for.
149
00:08:40.950 --> 00:08:45.190
Emerson Keenan: So the real difference here is that in
traditional methods we we have an idea about
150
00:08:45.190 --> 00:08:48.550
Emerson Keenan: how we think we want to extract the signal.
And an engineer kind of crafts an algorithm
151
00:08:48.550 --> 00:08:51.720
Emerson Keenan: based on the expert information they have,
whereas AI,
152
00:08:51.720 --> 00:08:54.640
Emerson Keenan: instead flips it on its head and says,
okay, we've got a signal,
153
00:08:54.640 --> 00:08:57.200
Emerson Keenan: we've got as an input and a signal we want to
get to as an output,
154
00:08:57.200 --> 00:09:00.080
Emerson Keenan: and we let the machine do the learning. So
essentially,
155
00:09:00.080 --> 00:09:04.160
Emerson Keenan: we've collected a very large data set of
recordings where we had both our device and
156
00:09:04.160 --> 00:09:08.280
Emerson Keenan: another technology on generally the
ultrasound based technology and measuring
157
00:09:08.280 --> 00:09:12.160
Emerson Keenan: them together, we can develop a model that
takes that input electrical signal and gives
158
00:09:12.160 --> 00:09:15.000
Emerson Keenan: us that output fetal heart rate with a much
higher accuracy.
159
00:09:15.160 --> 00:09:17.160
Emerson Keenan: So that's the first way we're using things.
160
00:09:17.160 --> 00:09:19.960
Emerson Keenan: And then the second lens,
which I was talking about is the ability to
161
00:09:19.960 --> 00:09:21.200
Emerson Keenan: do predictive analytics.
162
00:09:21.200 --> 00:09:24.680
Emerson Keenan: So not only can we predict those raw
physiological parameters,
163
00:09:24.680 --> 00:09:28.960
Emerson Keenan: but looking ahead to things like the
likelihood that women might go into labor.
164
00:09:28.960 --> 00:09:33.520
Emerson Keenan: And so there's things like the uterine
contractions that have also these electrical
165
00:09:33.520 --> 00:09:35.400
Emerson Keenan: patterns that exist in these signals.
166
00:09:35.400 --> 00:09:40.360
Emerson Keenan: So again feeding in that raw electrical data
and taking the outcomes that we know.
167
00:09:40.360 --> 00:09:42.720
Emerson Keenan: So if we follow this woman over time and work
out,
168
00:09:42.760 --> 00:09:46.440
Emerson Keenan: hey, the ones that went into delivery within
seven days had a certain kind of pattern.
169
00:09:46.480 --> 00:09:50.080
Emerson Keenan: Again, we can train the AI and better predict
these kind of outcomes.
170
00:09:50.080 --> 00:09:54.930
Emerson Keenan: And I think that's where AI is a really
transformative technology in that we can have
171
00:09:54.930 --> 00:09:57.610
Emerson Keenan: this expert level diagnosis and in some
cases,
172
00:09:57.610 --> 00:10:01.610
Emerson Keenan: even surpassing expert level diagnosis
available to anyone,
173
00:10:01.650 --> 00:10:03.410
Emerson Keenan: everyone, at any time.
174
00:10:03.410 --> 00:10:07.370
Emerson Keenan: And I think that's where it has these real
benefits, is having that really high accuracy
175
00:10:07.370 --> 00:10:09.090
Emerson Keenan: available at any time to the clinician.
176
00:10:09.410 --> 00:10:15.010
Sarath Ranganathan: In other bits of kit or a test that we use in
medicine we will refer to the positive
177
00:10:15.010 --> 00:10:17.770
Sarath Ranganathan: predictive value and negative predictive
value.
178
00:10:17.770 --> 00:10:20.090
Sarath Ranganathan: And look at receiver operating curves.
179
00:10:20.250 --> 00:10:25.890
Sarath Ranganathan: You're obviously using this device in
sometimes quite high risk situations or
180
00:10:25.890 --> 00:10:27.770
Sarath Ranganathan: situations that become high risk.
181
00:10:28.010 --> 00:10:31.170
Sarath Ranganathan: How confident are you in the performance of
the device?
182
00:10:31.530 --> 00:10:36.157
Sarath Ranganathan: Is it sort of comprehensive,
so it really does support the decision making
183
00:10:36.157 --> 00:10:40.810
Sarath Ranganathan: and I mean obstetrics is a fraught field so
can things go wrong?
184
00:10:41.330 --> 00:10:45.410
Emerson Keenan: Yeah, it's a fantastic question and a really
important thing we've looked at. So the
185
00:10:45.410 --> 00:10:49.660
Emerson Keenan: initial focus of technology providing that
data to the clinician means that they can
186
00:10:49.660 --> 00:10:53.060
Emerson Keenan: provide that clinical decision making
capacity in real time.
187
00:10:53.060 --> 00:10:56.900
Emerson Keenan: So the data is flowing in real time over the
internet to that clinician portal.
188
00:10:56.900 --> 00:11:00.260
Emerson Keenan: And they're looking at that data just like
they would if the patient was there in a
189
00:11:00.260 --> 00:11:04.220
Emerson Keenan: clinical setting. And so one of the real
valuable things we see is that sometimes we
190
00:11:04.260 --> 00:11:07.460
Emerson Keenan: find that women aren't able to attend these
appointments. So in these scenarios we don't
191
00:11:07.460 --> 00:11:10.940
Emerson Keenan: have the data at all. So if we can have this
ability to at least have that information
192
00:11:10.940 --> 00:11:13.300
Emerson Keenan: available, the clinician can then make that
decision.
193
00:11:13.300 --> 00:11:15.540
Emerson Keenan: And so if we do find something that's
concerning,
194
00:11:15.540 --> 00:11:17.740
Emerson Keenan: the doctor will say, hey,
we need you to come into hospital for further
195
00:11:17.740 --> 00:11:20.940
Emerson Keenan: monitoring or further checks. And so I think
that's a really important thing as part of
196
00:11:20.940 --> 00:11:24.260
Emerson Keenan: the clinical workflow. And then moving into
the predictive side of things,
197
00:11:24.540 --> 00:11:27.580
Emerson Keenan: the ability to have an AI,
predicting the outcomes,
198
00:11:27.580 --> 00:11:31.540
Emerson Keenan: we always want to be making sure that we are
better than what's currently available.
199
00:11:31.540 --> 00:11:34.940
Emerson Keenan: So, you know, there might be in the instance
of predicting the onset of labor.
200
00:11:34.940 --> 00:11:38.020
Emerson Keenan: Currently, there's a test called the fetal
fibronectin test,
201
00:11:38.420 --> 00:11:42.860
Emerson Keenan: which generally only has a positive
predictive value around 30 to 40%.
202
00:11:42.980 --> 00:11:47.580
Emerson Keenan: So if we can improve upon those statistics
through robust validation before this goes to
203
00:11:47.620 --> 00:11:49.990
Emerson Keenan: market showing we're greatly improving on
that.
204
00:11:50.070 --> 00:11:53.510
Emerson Keenan: That's really our goal to show that we can do
this better than the way things are doing
205
00:11:53.550 --> 00:11:54.190
Emerson Keenan: currently.
206
00:11:54.430 --> 00:11:59.070
Sarath Ranganathan: We're at the moment really striving to
connect people within this med school to
207
00:11:59.110 --> 00:12:04.630
Sarath Ranganathan: disciplines and professions and skill sets
outside of med school and other parts of
208
00:12:04.630 --> 00:12:08.230
Sarath Ranganathan: university, and more broadly in society
itself.
209
00:12:08.470 --> 00:12:13.910
Sarath Ranganathan: So I'm intrigued as to the process and the
conversations that enabled you to come into
210
00:12:13.910 --> 00:12:14.870
Sarath Ranganathan: our medical school.
211
00:12:14.910 --> 00:12:16.590
Emerson Keenan: Yeah, it's a really lovely story.
212
00:12:16.590 --> 00:12:21.070
Emerson Keenan: So very early in my PhD journey,
I said to my engineering supervisor,
213
00:12:21.110 --> 00:12:24.830
Emerson Keenan: you know, we need a clinician involved and
working through this with us.
214
00:12:24.870 --> 00:12:27.790
Emerson Keenan: You know, the only way we're going to
understand the clinical challenges and be
215
00:12:27.790 --> 00:12:31.190
Emerson Keenan: able to make this work is having a clinician
deeply embedded in the team.
216
00:12:31.190 --> 00:12:36.870
Emerson Keenan: And so he reached out to a doctor colleague
of his and was working in the neurology space
217
00:12:36.870 --> 00:12:40.030
Emerson Keenan: who, funnily enough, knew another obstetric
colleague,
218
00:12:40.190 --> 00:12:43.470
Emerson Keenan: who knew another obstetric colleague who had
the same idea as us,
219
00:12:43.470 --> 00:12:46.350
Emerson Keenan: which is now my co-founder and research
collaborator,
220
00:12:46.350 --> 00:12:47.600
Emerson Keenan: Doctor Fiona Brownfoot.
221
00:12:47.840 --> 00:12:50.880
Emerson Keenan: And so they kind of both said,
hey, you guys are thinking about a very
222
00:12:50.880 --> 00:12:54.560
Emerson Keenan: similar idea here. You should meet up and
talk further. And so we went out for a coffee
223
00:12:54.560 --> 00:12:58.440
Emerson Keenan: and started talking about the challenges with
with pregnancy monitoring and realized very
224
00:12:58.440 --> 00:13:01.920
Emerson Keenan: quickly we were on the same page. We both
wanted to solve the same problem just from
225
00:13:01.920 --> 00:13:05.920
Emerson Keenan: the opposite lens. We had an engineering
solution and they had that clinical need.
226
00:13:05.920 --> 00:13:09.520
Emerson Keenan: And I think it's such an important part of
developing technology that's actually
227
00:13:09.520 --> 00:13:13.400
Emerson Keenan: valuable for the clinical setting,
is having that clinical guidance really early
228
00:13:13.400 --> 00:13:18.680
Emerson Keenan: in the process. And I think the more we can
do as a school to enable that will be really
229
00:13:18.680 --> 00:13:21.760
Emerson Keenan: important. And I think from there,
it was interesting for me as an engineer,
230
00:13:21.800 --> 00:13:25.720
Emerson Keenan: working with Fiona throughout the course of
my PhD and just getting more and more
231
00:13:25.720 --> 00:13:29.800
Emerson Keenan: interested in that clinical application,
that I actually moved over to the medical
232
00:13:29.800 --> 00:13:33.280
Emerson Keenan: school. So I'm now an honorary research
fellow. I spent a bit of time as a postdoc in
233
00:13:33.320 --> 00:13:36.480
Emerson Keenan: the medical facility,
and it's for me, it's just so exciting to be
234
00:13:36.480 --> 00:13:39.280
Emerson Keenan: embedded in that clinical space. You know,
working in the clinic,
235
00:13:39.280 --> 00:13:42.360
Emerson Keenan: seeing how the technology is used. And it's
interesting as an engineer, often they're not
236
00:13:42.360 --> 00:13:44.920
Emerson Keenan: used how you think they are,
which is so important to see.
237
00:13:44.960 --> 00:13:47.680
Emerson Keenan: You might design it a certain way,
but in the reality of it,
238
00:13:47.680 --> 00:13:49.600
Emerson Keenan: things get used in the clinical workflows.
239
00:13:49.600 --> 00:13:53.000
Emerson Keenan: They need to be quick. They need to be easy
to understand. I think the more things we can
240
00:13:53.000 --> 00:13:57.280
Emerson Keenan: do to bring together engineers who have an
idea and clinicians have a problem,
241
00:13:57.680 --> 00:13:59.800
Emerson Keenan: it's going to be really great to see how that
all evolves.
242
00:13:59.960 --> 00:14:03.760
Sarath Ranganathan: I was going to ask you how you perceive
clinicians to think,
243
00:14:03.920 --> 00:14:07.040
Sarath Ranganathan: and you sort of summarized it there that
clinicians have just got problems.
244
00:14:07.080 --> 00:14:10.360
Emerson Keenan: Yeah, they've got problems and not enough
time. And you need to make it very easy to
245
00:14:10.400 --> 00:14:12.840
Emerson Keenan: solve the problem quickly and don't make it
hard.
246
00:14:13.040 --> 00:14:16.320
Emerson Keenan: If there's a 100 page instruction manual and
you've got to do 20 steps, it's just not
247
00:14:16.320 --> 00:14:19.000
Emerson Keenan: going to work. It's got to be quick,
easy and efficient to use.
248
00:14:19.040 --> 00:14:20.360
Sarath Ranganathan: Yeah, I would agree with that.
249
00:14:20.680 --> 00:14:24.360
Sarath Ranganathan: Now people often refer to the the black box
of algorithms,
250
00:14:24.360 --> 00:14:26.320
Sarath Ranganathan: particularly in predictive algorithms.
251
00:14:26.640 --> 00:14:31.040
Sarath Ranganathan: And there are commercial devices like on my
watch it can measure my heart rate.
252
00:14:31.320 --> 00:14:34.280
Sarath Ranganathan: Do you as an electrical engineer interested
in this area,
253
00:14:34.480 --> 00:14:40.760
Sarath Ranganathan: have some deep insights into how exactly it's
monitoring the heart rate there on my watch,
254
00:14:40.800 --> 00:14:46.970
Sarath Ranganathan: and therefore, are there any similarities in
that technology or the use case with you and
255
00:14:46.970 --> 00:14:49.010
Sarath Ranganathan: your device that can measure fetal heart
rate?
256
00:14:49.050 --> 00:14:52.410
Emerson Keenan: Yeah, it's absolutely fascinating,
the stuff that the smartwatch companies are
257
00:14:52.410 --> 00:14:56.290
Emerson Keenan: bringing out with these predictive analytics
and the data sets they've collected are
258
00:14:56.450 --> 00:15:00.690
Emerson Keenan: enormous. Like, I think I've seen an Apple
Watch study that was at least 50 or 60,000
259
00:15:00.690 --> 00:15:04.650
Emerson Keenan: participants looking at atrial fibrillation.
So there's a few elements to the
260
00:15:04.650 --> 00:15:07.170
Emerson Keenan: explainability. You know,
in the early days of AI,
261
00:15:07.570 --> 00:15:10.090
Emerson Keenan: I think we were just focusing on the
performance benefits. Just saying, hey, how
262
00:15:10.090 --> 00:15:13.770
Emerson Keenan: can we get the best performance possible? But
there has been a real shift to move to more
263
00:15:13.770 --> 00:15:16.770
Emerson Keenan: explainable AI techniques,
and there's been a lot of good research done
264
00:15:16.770 --> 00:15:19.890
Emerson Keenan: in how we can better understand what these
models are picking up on.
265
00:15:19.890 --> 00:15:24.450
Emerson Keenan: And so we've made methods that we can
actually project back onto the raw signals,
266
00:15:24.610 --> 00:15:26.770
Emerson Keenan: the things that the AI models are detecting.
267
00:15:27.370 --> 00:15:30.610
Emerson Keenan: So I'll talk briefly about our own use case
first and then move into,
268
00:15:30.650 --> 00:15:34.370
Emerson Keenan: I think, how the smartwatch might evolve in
the CTG analysis.
269
00:15:34.410 --> 00:15:38.850
Emerson Keenan: We actually published a study recently,
which we did this back propagation to show
270
00:15:38.890 --> 00:15:42.090
Emerson Keenan: what the AI was picking up,
and it was really fascinating without any
271
00:15:42.090 --> 00:15:47.340
Emerson Keenan: guidance or training. When we project the
features that are associated with fetal
272
00:15:47.340 --> 00:15:51.180
Emerson Keenan: distress. A lot of them were very similar
features to what clinicians associate with
273
00:15:51.180 --> 00:15:54.820
Emerson Keenan: fetal distress. So without any guidance on
what the physiology was or anything like
274
00:15:54.820 --> 00:15:59.100
Emerson Keenan: that, these highlights in the signals aligned
very closely with clinical guidance.
275
00:15:59.100 --> 00:16:02.300
Emerson Keenan: And so it was really interesting for us to
see that these models do pick up some of the
276
00:16:02.300 --> 00:16:05.220
Emerson Keenan: same things we've learned throughout years
and centuries,
277
00:16:05.220 --> 00:16:07.180
Emerson Keenan: in some cases of studying these signals.
278
00:16:07.180 --> 00:16:08.980
Emerson Keenan: So I think that's really interesting.
279
00:16:09.020 --> 00:16:12.540
Emerson Keenan: You know, seeing that reflected back to the
clinician when they go, hey, you know, some
280
00:16:12.540 --> 00:16:15.660
Emerson Keenan: of the things that's detecting are similar.
What I'm detecting is just highlighting them
281
00:16:15.660 --> 00:16:19.300
Emerson Keenan: for me and showing me really clearly what's
leading to that decision that's making. So I
282
00:16:19.300 --> 00:16:23.060
Emerson Keenan: think bringing explainable AI into these
methods is really important to get that
283
00:16:23.060 --> 00:16:24.980
Emerson Keenan: clinician confidence on what's going on.
284
00:16:25.380 --> 00:16:30.020
Emerson Keenan: In terms of the smartwatch side of things,
there's always a limitation around where
285
00:16:30.060 --> 00:16:32.540
Emerson Keenan: sensors are placed and what they can actually
pick up.
286
00:16:32.540 --> 00:16:37.420
Emerson Keenan: So I would love if we could have a pure
smartwatch that could measure everything we
287
00:16:37.420 --> 00:16:41.380
Emerson Keenan: need. But unfortunately the fetal signals are
very located to the abdomen.
288
00:16:41.380 --> 00:16:45.110
Emerson Keenan: So I think in this case they'll always need
to be some sort of abdominal based sensor
289
00:16:45.110 --> 00:16:47.070
Emerson Keenan: that measures that fetal component.
290
00:16:47.430 --> 00:16:50.950
Emerson Keenan: But I think the smartwatch does have a lot of
interesting applications looking at the
291
00:16:50.950 --> 00:16:53.030
Emerson Keenan: maternal side of things. Maternal heart rate.
292
00:16:53.390 --> 00:16:56.150
Emerson Keenan: There's been some early studies showing
association with maternal heart rate and
293
00:16:56.150 --> 00:17:00.030
Emerson Keenan: onset to delivery. So I think these platforms
are going to be really interesting as they
294
00:17:00.030 --> 00:17:03.030
Emerson Keenan: evolve. Having your smartwatch picking up so
much information,
295
00:17:03.030 --> 00:17:06.710
Emerson Keenan: and I think feeding that back to your doctor
and having that available is the next
296
00:17:06.710 --> 00:17:10.070
Emerson Keenan: frontier of clinical care. You know,
integrating these insights into all sorts of
297
00:17:10.070 --> 00:17:13.190
Emerson Keenan: medicine is somewhere,
I think, you know, both both I know the
298
00:17:13.190 --> 00:17:15.990
Emerson Keenan: companies themselves are looking into. But we
should think about more broadly how we
299
00:17:15.990 --> 00:17:17.630
Emerson Keenan: integrate this into medical practice.
300
00:17:17.670 --> 00:17:20.350
Sarath Ranganathan: Now I've got what potentially could be a dumb
question.
301
00:17:20.350 --> 00:17:23.630
Sarath Ranganathan: And if you think it's a dumb question,
you don't have to answer.
302
00:17:23.630 --> 00:17:28.430
Sarath Ranganathan: But I do recall a study many years ago was
done by a Swiss colleague of mine that looked
303
00:17:28.430 --> 00:17:32.630
Sarath Ranganathan: at exacerbations of asthma and showed that
they were predictable,
304
00:17:32.630 --> 00:17:34.310
Sarath Ranganathan: although in a very unusual way.
305
00:17:34.310 --> 00:17:37.550
Sarath Ranganathan: So the mathematics that predicted it was
fractals.
306
00:17:37.550 --> 00:17:39.110
Sarath Ranganathan: So sort of chaos theory.
307
00:17:39.110 --> 00:17:44.200
Sarath Ranganathan: And the question I have is and I've asked
this of other AI experts is are the
308
00:17:44.200 --> 00:17:50.160
Sarath Ranganathan: algorithms when you actually analyze them,
are they sort of fractal type mathematics or
309
00:17:50.160 --> 00:17:53.480
Sarath Ranganathan: does anyone know or is it really genuinely a
black box?
310
00:17:54.560 --> 00:17:56.680
Emerson Keenan: Oh, it's a it's a tough question.
311
00:17:57.440 --> 00:18:00.080
Emerson Keenan: It depends on the type of model you're
training.
312
00:18:00.080 --> 00:18:04.600
Emerson Keenan: So different model structures will pick up on
different kinds of information at a very high
313
00:18:04.600 --> 00:18:08.440
Emerson Keenan: level. You know, you might have things like a
convolutional network with a certain window
314
00:18:08.440 --> 00:18:10.680
Emerson Keenan: size, and they will only look at a very,
you know,
315
00:18:10.720 --> 00:18:12.600
Emerson Keenan: restricted piece of the signal.
316
00:18:12.600 --> 00:18:16.600
Emerson Keenan: Whereas some other model architectures like
your LSTMs or a transformer model,
317
00:18:16.600 --> 00:18:19.520
Emerson Keenan: do look at these long range patterns and
dependencies.
318
00:18:19.520 --> 00:18:23.480
Emerson Keenan: So that might be somewhere where the model
structure itself can determine what you're
319
00:18:23.520 --> 00:18:25.080
Emerson Keenan: actually picking up in the signal.
320
00:18:25.080 --> 00:18:27.600
Emerson Keenan: So I think there is always a bit of,
you know,
321
00:18:27.640 --> 00:18:31.400
Emerson Keenan: human nous, or at least currently to know
what the best kind of model structure is for
322
00:18:31.400 --> 00:18:34.800
Emerson Keenan: a given task and knowing the structure of the
physiological signal.
323
00:18:34.800 --> 00:18:38.480
Emerson Keenan: So I think it's interesting. I think these
things knowing the fractal dynamic can guide
324
00:18:38.480 --> 00:18:42.530
Emerson Keenan: you then to make the right AI model decision
to get the best result.
325
00:18:42.930 --> 00:18:44.090
Sarath Ranganathan: Emerson, you're so polite.
326
00:18:44.130 --> 00:18:47.170
Sarath Ranganathan: You said that was a tough question. I don't
know if that's another way,
327
00:18:47.210 --> 00:18:49.450
Sarath Ranganathan: politely, of saying it was actually a dumb
question,
328
00:18:49.450 --> 00:18:57.130
Sarath Ranganathan: but. But I've come across you before and in
that meeting where I met you,
329
00:18:57.130 --> 00:18:59.210
Sarath Ranganathan: you were giving a wonderful presentation.
330
00:18:59.210 --> 00:19:02.570
Sarath Ranganathan: You had explored AI technologies for
research.
331
00:19:02.570 --> 00:19:06.370
Sarath Ranganathan: And I think you've also done a little bit in
terms of other aspects of healthcare related
332
00:19:06.370 --> 00:19:11.155
Sarath Ranganathan: to Kali, etc.. How do you see,
because I see you as an expert here,
333
00:19:11.155 --> 00:19:15.570
Sarath Ranganathan: how do you see AI technologies continuing to
evolve in healthcare?
334
00:19:15.570 --> 00:19:17.650
Sarath Ranganathan: Can you predict what's going to happen?
335
00:19:18.130 --> 00:19:21.130
Emerson Keenan: I definitely can't predict it,
but I think there's some there's some early
336
00:19:21.130 --> 00:19:22.850
Emerson Keenan: signs about where things might go.
337
00:19:23.050 --> 00:19:25.970
Emerson Keenan: And I'm really excited by the potentials that
all of this is bringing.
338
00:19:26.290 --> 00:19:30.170
Emerson Keenan: You know, some of the techniques we've been
working on have been around for a while.
339
00:19:30.170 --> 00:19:33.810
Emerson Keenan: This ability to do the taking in input data
and predicting what,
340
00:19:33.810 --> 00:19:38.130
Emerson Keenan: what might come next. But the emergence of
generative AI techniques and large language
341
00:19:38.130 --> 00:19:42.740
Emerson Keenan: models has just really unlocked a huge area
of exploration.
342
00:19:42.740 --> 00:19:46.100
Emerson Keenan: I think one of the things that's most
exciting to me is the area of scientific
343
00:19:46.100 --> 00:19:49.820
Emerson Keenan: discovery. You know, the ability of these
tools to bring in large,
344
00:19:49.820 --> 00:19:53.980
Emerson Keenan: large amounts of information and really
concisely pick out the things that might be
345
00:19:53.980 --> 00:19:56.980
Emerson Keenan: really important. And I think one of the
things I've mentioned in a few of my
346
00:19:56.980 --> 00:20:01.900
Emerson Keenan: presentations is Google recently developed
what they call an AI code scientist,
347
00:20:02.060 --> 00:20:06.940
Emerson Keenan: and the idea of this model was that we can
give it some indication of an area we might
348
00:20:06.940 --> 00:20:10.380
Emerson Keenan: want to look into further,
and it goes away and generates hypotheses,
349
00:20:10.420 --> 00:20:13.940
Emerson Keenan: compares those hypotheses against data that's
already in the literature, and kind of does
350
00:20:13.940 --> 00:20:18.340
Emerson Keenan: this continual ranking to kind of narrow down
what might be the most attractive hypothesis
351
00:20:18.340 --> 00:20:20.140
Emerson Keenan: based on everything it's looked at.
352
00:20:20.140 --> 00:20:23.700
Emerson Keenan: And some of those early studies I know they
were looking at trying to find a drug
353
00:20:23.700 --> 00:20:28.980
Emerson Keenan: repurposing for a novel tumor candidate,
and they ran the the AI code scientists,
354
00:20:28.980 --> 00:20:30.660
Emerson Keenan: and it came up with a candidate.
355
00:20:30.660 --> 00:20:33.780
Emerson Keenan: And the scientists then went and validated
that there was nothing in the literature that
356
00:20:33.780 --> 00:20:37.260
Emerson Keenan: had ever used this drug candidate for that
purpose before, but then it validated,
357
00:20:37.300 --> 00:20:39.670
Emerson Keenan: they found that it could treat that tumor.
So, you know,
358
00:20:39.710 --> 00:20:42.830
Emerson Keenan: these kind of things where we,
you know, set the AI in a certain direction.
359
00:20:42.830 --> 00:20:45.550
Emerson Keenan: We say, hey, we've got this really
interesting problem space. I want you to go
360
00:20:45.550 --> 00:20:49.470
Emerson Keenan: and look at everything that's out there and
tell me what's the most promising candidate.
361
00:20:49.510 --> 00:20:50.830
Emerson Keenan: I'm just I'm just so excited.
362
00:20:50.830 --> 00:20:54.230
Emerson Keenan: The ability that we could narrow down these
things really quickly to know where we need
363
00:20:54.230 --> 00:20:57.150
Emerson Keenan: to go. And I think that's just such an
exciting area to move into.
364
00:20:57.430 --> 00:20:59.550
Sarath Ranganathan: With that positive example in mind.
365
00:20:59.590 --> 00:21:04.070
Sarath Ranganathan: What advice would you give to early career
researchers who are working or anticipating
366
00:21:04.070 --> 00:21:05.870
Sarath Ranganathan: working in AI or digital health?
367
00:21:05.910 --> 00:21:11.470
Emerson Keenan: The main advice I give when when I hear this
question is I think AI is coming.
368
00:21:11.470 --> 00:21:15.190
Emerson Keenan: It's much like the internet,
smartphones, these tools.
369
00:21:15.390 --> 00:21:18.310
Emerson Keenan: You know, there was this early days when a
few people were kind of using them and
370
00:21:18.310 --> 00:21:21.550
Emerson Keenan: adopting them, and all of a sudden we look
around the world, they're everywhere. They're
371
00:21:21.590 --> 00:21:22.990
Emerson Keenan: a necessity. Part of doing work.
372
00:21:22.990 --> 00:21:24.790
Emerson Keenan: You can't do work without using them.
373
00:21:24.790 --> 00:21:28.190
Emerson Keenan: So I think we need to treat it in the same
way that they are going to become part of the
374
00:21:28.190 --> 00:21:31.150
Emerson Keenan: way we do work, and we just need to start
using them,
375
00:21:31.150 --> 00:21:35.070
Emerson Keenan: start learning them, adopting our practices,
because the ones who do adopt them are going
376
00:21:35.070 --> 00:21:37.230
Emerson Keenan: to have an unfair advantage on those that
don't.
377
00:21:37.550 --> 00:21:41.750
Emerson Keenan: And the sooner you can get into the practice
of using them and working out how they work
378
00:21:41.750 --> 00:21:43.430
Emerson Keenan: will make your practice a lot easier.
379
00:21:43.430 --> 00:21:46.710
Emerson Keenan: And I think some of the the guiding things I
would say in doing this are,
380
00:21:46.990 --> 00:21:49.590
Emerson Keenan: you know, when when you go to use a new AI
technique,
381
00:21:49.710 --> 00:21:52.590
Emerson Keenan: think very carefully about the problem you
want to apply it for.
382
00:21:52.630 --> 00:21:57.150
Emerson Keenan: First, think about all the problems you might
use it on and find one that really is
383
00:21:57.150 --> 00:21:59.110
Emerson Keenan: amenable to being solved using AI.
384
00:21:59.150 --> 00:22:01.870
Emerson Keenan: Don't just have AI and go,
hey, I'm going to put it on every single
385
00:22:01.870 --> 00:22:05.030
Emerson Keenan: problem. I know it's finding where it's most
suited to that problem.
386
00:22:05.030 --> 00:22:08.390
Emerson Keenan: So I think really defining that really
clearly upfront is really important.
387
00:22:08.390 --> 00:22:11.390
Emerson Keenan: And I think things like the community of
practice. I know there's a community of
388
00:22:11.390 --> 00:22:14.710
Emerson Keenan: practice here at the university looking at
how we can use these AI techniques, you know,
389
00:22:14.750 --> 00:22:18.830
Emerson Keenan: developing these things in your lab groups,
in your your friendships and collegial
390
00:22:18.830 --> 00:22:22.270
Emerson Keenan: networks on the best practice of using these
things. And the more you can do that,
391
00:22:22.310 --> 00:22:26.630
Emerson Keenan: the faster ahead of the curve you will be as
they become mainstay until you know they are
392
00:22:26.670 --> 00:22:28.310
Emerson Keenan: just a way we do our work.
393
00:22:28.590 --> 00:22:31.550
Sarath Ranganathan: So I've got another tough replica question.
394
00:22:31.950 --> 00:22:35.910
Sarath Ranganathan: What advice would you give to the head of the
medical school about how we should be
395
00:22:36.080 --> 00:22:39.320
Sarath Ranganathan: approaching AI or accommodating developments
in AI?
396
00:22:39.840 --> 00:22:41.640
Emerson Keenan: So it's an interesting one,
I think.
397
00:22:41.680 --> 00:22:45.920
Emerson Keenan: I think encouraging this practice and making
it easier for people to access these tools is
398
00:22:45.920 --> 00:22:49.240
Emerson Keenan: really important. I know only the university
has an internal platform, Spark,
399
00:22:49.440 --> 00:22:51.080
Emerson Keenan: where researchers can can use this.
400
00:22:51.080 --> 00:22:54.200
Emerson Keenan: So I think really prioritizing access to
these tools,
401
00:22:54.520 --> 00:22:56.600
Emerson Keenan: training on these tools is really important.
402
00:22:56.760 --> 00:23:00.280
Emerson Keenan: I'm not sure if this is 100% the way to go,
but I know some of the big companies even
403
00:23:00.280 --> 00:23:03.040
Emerson Keenan: have leaderboards on who is using these tools
the most,
404
00:23:03.040 --> 00:23:06.320
Emerson Keenan: and having bonuses and prizes for who's
adopting these tools the most.
405
00:23:06.320 --> 00:23:10.120
Emerson Keenan: So that's an interesting way of doing things.
You know, that way you're incentivizing the
406
00:23:10.120 --> 00:23:13.360
Emerson Keenan: use of these tools. I think it could be
something really interesting to explore.
407
00:23:13.400 --> 00:23:15.520
Sarath Ranganathan: Well, I can certainly afford a virtual prize,
408
00:23:15.920 --> 00:23:18.040
Sarath Ranganathan: so I use AI in Spotify.
409
00:23:18.200 --> 00:23:20.120
Sarath Ranganathan: It chooses some of the songs I'd like.
410
00:23:20.160 --> 00:23:21.600
Sarath Ranganathan: I presume that's that's AI.
411
00:23:21.640 --> 00:23:26.040
Sarath Ranganathan: So I'm going to move on. Now to the final
question is what are you listening to right
412
00:23:26.040 --> 00:23:31.520
Sarath Ranganathan: now, Emerson? And how does it relate to your
work and your life experiences?
413
00:23:31.560 --> 00:23:32.680
Sarath Ranganathan: That's a complex question.
414
00:23:32.720 --> 00:23:37.675
Emerson Keenan: Interesting. So actually I'm listening to the
newest album by a band called Geese lately,
415
00:23:37.675 --> 00:23:42.610
Emerson Keenan: like the bird. And the song I really like off
that album is a song called Husbands.
416
00:23:43.250 --> 00:23:45.810
Emerson Keenan: It's difficult to explain without listening
to the track itself, but it's super
417
00:23:45.810 --> 00:23:48.330
Emerson Keenan: interesting melodically and rhythmically.
418
00:23:48.330 --> 00:23:51.850
Emerson Keenan: So there's, you know,
all these different guitar parts happening.
419
00:23:51.890 --> 00:23:55.770
Emerson Keenan: There's a very interesting drum line,
and it's all kind of weaved together in this
420
00:23:55.770 --> 00:23:59.570
Emerson Keenan: really complex way. And I think,
I think it's really beautiful when,
421
00:23:59.570 --> 00:24:03.250
Emerson Keenan: when all these things can come together in a
really cohesive way.
422
00:24:03.250 --> 00:24:05.970
Emerson Keenan: And I think bringing together,
you know, different skill sets,
423
00:24:06.010 --> 00:24:08.730
Emerson Keenan: engineering, science,
medicine, the human element,
424
00:24:08.730 --> 00:24:12.530
Emerson Keenan: I think is really resonant of what we should
be trying to do in the medical school and in
425
00:24:12.530 --> 00:24:16.530
Emerson Keenan: biomedical research more generally is how do
we bring all these tools together?
426
00:24:16.570 --> 00:24:20.690
Emerson Keenan: Not forgetting the human element,
you know, thinking about how it makes us do
427
00:24:20.690 --> 00:24:23.770
Emerson Keenan: our jobs better, live our lives better,
and you know,
428
00:24:23.810 --> 00:24:24.890
Emerson Keenan: develop a better world.
429
00:24:25.290 --> 00:24:27.130
Sarath Ranganathan: Well, it's been lovely talking to you.
430
00:24:27.170 --> 00:24:29.130
Sarath Ranganathan: Congratulations on Kali Healthcare.
431
00:24:29.130 --> 00:24:31.650
Sarath Ranganathan: I hope that you have every success with it.
432
00:24:31.770 --> 00:24:37.300
Sarath Ranganathan: I think you're a great example of you know,
involving someone from outside the medical or
433
00:24:37.300 --> 00:24:38.820
Sarath Ranganathan: health disciplines in health.
434
00:24:38.820 --> 00:24:41.700
Sarath Ranganathan: And just to see what can be achieved through
that,
435
00:24:41.700 --> 00:24:45.900
Sarath Ranganathan: I think can have a wonderful future in our
school. I hope you stay within our school and
436
00:24:45.940 --> 00:24:49.500
Sarath Ranganathan: that we build a community of people with a
similar interests as you.
437
00:24:49.500 --> 00:24:53.380
Sarath Ranganathan: So thanks for your time and I look forward to
listening to Geese.
438
00:24:53.740 --> 00:24:55.460
Emerson Keenan: Thanks. It's been a pleasure being on the
show.
439
00:24:57.020 --> 00:24:59.300
Sarath Ranganathan: Thanks for joining us on today's episode.
440
00:24:59.900 --> 00:25:03.020
Sarath Ranganathan: If you have a question or a brilliant idea
for a future topic,
441
00:25:03.020 --> 00:25:06.140
Sarath Ranganathan: we'd love to hear from you via the link in
the show notes.
442
00:25:06.860 --> 00:25:10.220
Sarath Ranganathan: And feel free to share this episode with a
friend or colleague,
443
00:25:10.380 --> 00:25:11.980
Sarath Ranganathan: or even a member of your family.
444
00:25:12.460 --> 00:25:13.500
Sarath Ranganathan: See you next time!
