النص 🔗
1
00:00:02,200 --> 00:00:03,700
Welcome to the Circuit Python Show.
2
00:00:03,980 --> 00:00:05,100
I'm your host, Paul Cutler.
3
00:00:05,580 --> 00:00:08,760
This episode, I'm joined by Michelle Hui and Reitwiec Shandilya.
4
00:00:09,640 --> 00:00:13,580
Michelle is a master's student at Cornell Tech and AI and Urban Technology.
5
00:00:14,180 --> 00:00:18,080
Her practice includes accessible technology for good across health and climate.
6
00:00:18,620 --> 00:00:23,260
Previously, she has worked at Wing, Google's drone delivery moonshot, and the United Nations.
7
00:00:24,200 --> 00:00:28,320
Reitwiec is a master's student at Cornell Tech and a fellow at 645 Ventures,
8
00:00:28,700 --> 00:00:33,120
where he focuses on early stage investments in AI infrastructure, security, and hardware.
9
00:00:33,720 --> 00:00:39,000
Previously, he was an early member of SAP's AI team working on Enterprise AI Solutions.
10
00:00:41,100 --> 00:00:42,640
Michelle and Reitwiec, welcome to the show.
11
00:00:43,060 --> 00:00:44,840
Thank you for having us, so excited to be on.
12
00:00:45,360 --> 00:00:46,180
Yeah, it's great to be here Paul.
13
00:00:47,060 --> 00:00:51,080
You've worked together to develop the open pressure sensor device or open PSD.
14
00:00:51,520 --> 00:00:52,660
What is the open PSD?
15
00:00:53,460 --> 00:00:57,800
So the open PSD, we began working with Weill Cornell - New York Presbyterian.
16
00:00:58,360 --> 00:01:03,800
But after, you know, female patients have cancer, specific breast cancer and they get mastectomy,
17
00:01:04,720 --> 00:01:10,900
it leads to a serious loss of quality of life because they often say that their chest feels like it doesn't belong to themselves anymore.
18
00:01:11,160 --> 00:01:13,160
When they hug their children, they can't feel their children.
19
00:01:13,960 --> 00:01:21,160
So there's a surgeon at Wilde doing super cutting edge research where he essentially can reconnect your nerves and help you regain sensation in your chest.
20
00:01:21,680 --> 00:01:24,720
But they had been using this proprietary medical device.
21
00:01:25,400 --> 00:01:27,800
And the private company decided that it was no longer profitable
22
00:01:27,840 --> 00:01:33,040
anymore and they essentially abandoned device. And not just this surgeon, but essentially all the
23
00:01:33,240 --> 00:01:38,980
research labs across the U.S. doing any sort of nerve sensation, whether that's diabetes and
24
00:01:39,040 --> 00:01:43,740
you do sensation in your foot or your fingers, essentially lost access to this one device.
25
00:01:44,440 --> 00:01:49,260
So what we did was we actually engineered that from Blackbox and we rebuilt this pressure
26
00:01:49,400 --> 00:01:53,480
sensor device, hence the name, open PSD, open pressure sensor device.
27
00:01:54,380 --> 00:01:57,740
Well, yeah, I go with this was essentially to quantify sensation.
28
00:01:57,780 --> 00:02:04,080
and their function and help expand it to other clinical and neuroticetics as Michelle's
29
00:02:05,030 --> 00:02:09,420
what were some of the problem gaps that you were trying to resolve in addition to that
30
00:02:10,899 --> 00:02:16,760
I think beyond just this proprietary medical device before that they literally used to use one
31
00:02:17,040 --> 00:02:22,420
monofilament and you would just use a filament to poke at various areas of your skin so it's the super
32
00:02:22,720 --> 00:02:27,720
outdated archaic method and so our open source device isn't just replacing a abandonment
33
00:02:27,740 --> 00:02:32,860
medical device, but it's actually innovating on this super archaic nerve sensation method.
34
00:02:33,540 --> 00:02:37,620
And on top of that, we're also working with OSHWA, the Open Source Hardware Association,
35
00:02:38,240 --> 00:02:44,040
to explore, you know, what happens to any medical device, when expensive med tech decides
36
00:02:44,180 --> 00:02:49,820
that it's just not profitable in pain anymore, how can we democratize healthcare medical devices,
37
00:02:50,040 --> 00:02:52,920
how can we make a framework for open source, but also reliable?
38
00:02:53,640 --> 00:02:57,480
And so those were the two, I think, larger impacts that we're trying to address with the project.
39
00:02:58,640 --> 00:03:01,960
How did you collaborate with clinicians during the design process?
40
00:03:03,560 --> 00:03:08,840
Something that we discovered when we saw the proprietary device was that had been completely over-engineered.
41
00:03:09,660 --> 00:03:10,620
It had been super expensive.
42
00:03:10,860 --> 00:03:17,320
They came with this giant briefcase, you know, almost Matrix-style supervillain-style briefcase.
43
00:03:17,940 --> 00:03:24,960
And they pulled out like a proprietary tablet, almost their own iPad, and it had their software installed,
44
00:03:25,200 --> 00:03:27,240
And then they had this giant acquisition unit.
45
00:03:27,520 --> 00:03:30,600
And then they had this, like, this super boxy handheld device.
46
00:03:32,120 --> 00:03:34,800
And we just felt like it had been so over-engineered.
47
00:03:34,920 --> 00:03:37,760
And we could tell that they're probably trying to upsell it, you know, have software
48
00:03:38,000 --> 00:03:38,180
maintenance.
49
00:03:39,240 --> 00:03:42,780
And so we came in and we asked them even on the first day, like, if you could get rid of all of this,
50
00:03:42,860 --> 00:03:44,140
what would you want to get rid of, right?
51
00:03:45,480 --> 00:03:49,800
And we were able to change that, all of that giant briefcase into literally the size
52
00:03:49,860 --> 00:03:50,240
of a screwdriver.
53
00:03:50,640 --> 00:03:51,360
And that's all you needed.
54
00:03:51,380 --> 00:03:51,960
It was a screwdriver.
55
00:03:52,320 --> 00:03:55,160
And it was, you know, software that you could install in your own Mac.
56
00:03:55,180 --> 00:03:57,300
And so we work in the clinicians.
57
00:03:57,840 --> 00:04:00,240
Every single couple weeks, they would
Load full transcript
come back and give us feedback.
58
00:04:00,980 --> 00:04:06,740
I remember one time the surgeon came in and he felt our device and he's like, oh, your wires
59
00:04:07,120 --> 00:04:10,200
that are doing the sentencing need to adjust because it's actually hitting other areas
60
00:04:10,300 --> 00:04:10,740
in the skin.
61
00:04:11,360 --> 00:04:14,480
And it was just something we had totally, you know, not thought about and we didn't understand
62
00:04:14,580 --> 00:04:15,780
how they were using it at all.
63
00:04:16,880 --> 00:04:20,840
And so I think what we're able to do in open source and what we discovered is that because
64
00:04:21,060 --> 00:04:25,140
open source isn't meant to make money, you don't have to try to.
65
00:04:25,160 --> 00:04:30,080
start upselling everyone and you don't have to create this black box med tech industry,
66
00:04:30,250 --> 00:04:34,700
we can actually go back and forth with the clinicians and really improve and do exactly
67
00:04:35,000 --> 00:04:40,080
participatory design and create a participatory framework to improve like clinician efficiency
68
00:04:40,680 --> 00:04:42,460
and just improve the workflow overall.
69
00:04:43,200 --> 00:04:47,000
And something really exciting here was that, I mean, I don't know, we had really thought
70
00:04:47,100 --> 00:04:52,140
about before was the initial device that they came up with was, I mean, it looked really
71
00:04:52,140 --> 00:04:56,060
daunting as well. I think as a patient, you're really in a very vulnerable state when you're
72
00:04:56,100 --> 00:05:00,580
going through this entire process. Imagine someone coming in with a huge briefcase and takes out
73
00:05:00,580 --> 00:05:09,800
this huge gigantic device. It's really, really unnerving sometimes. And so our whole goal there
74
00:05:09,950 --> 00:05:15,540
as well was we want to make patients safe, you know, we want to trust that device.
75
00:05:16,620 --> 00:05:22,060
And I think since since it has gone to the clinic, we're still waiting for
76
00:05:22,120 --> 00:05:25,680
for the PI's feedback, but we're looking forward to them.
77
00:05:26,560 --> 00:05:28,660
You mentioned that you got it down to the size of a screw.
78
00:05:29,120 --> 00:05:30,640
Tell me a little bit more about the hardware,
79
00:05:31,120 --> 00:05:34,340
and then tell me what role Circuit Python played in the development of the firmware.
80
00:05:36,000 --> 00:05:37,960
CircuitPython was our firmware,
81
00:05:39,040 --> 00:05:43,160
and the idea, the way it was we had a load cell,
82
00:05:43,760 --> 00:05:45,520
and we have filaments attached to it,
83
00:05:45,720 --> 00:05:49,760
and we had sort of custom-wired and custom-twisted this filament at the end of the load-cell.
84
00:05:50,480 --> 00:05:54,640
And this load cell then had a sensor, a pressure sensor, which is attached to the load cell,
85
00:05:54,720 --> 00:05:59,440
which then goes to our firmware where we processed these signals, and we had some custom logic.
86
00:06:00,000 --> 00:06:05,100
Just for getting, let's say, set a ton of values, taking an average out of it.
87
00:06:05,480 --> 00:06:10,020
And whenever someone presses the button, it calculates a set pressure that's being applied to the load cell.
88
00:06:10,930 --> 00:06:16,440
And from there, we display on a UI that we had created, which is running locally as well,
89
00:06:16,540 --> 00:06:20,300
because we want to keep it on cloud, but it was personal patient data.
90
00:06:20,380 --> 00:06:20,740
as well.
91
00:06:22,200 --> 00:06:25,620
But yeah, that's how the overall device was set up in the hardware.
92
00:06:27,280 --> 00:06:29,700
What kind of screen was the UI displayed upon?
93
00:06:30,780 --> 00:06:31,400
It was just a laptop.
94
00:06:31,470 --> 00:06:35,520
It was also, I mean, we just coded up and React.
95
00:06:35,970 --> 00:06:42,220
And I think, so the other goal that we want to see with this project was how quickly can we actually get feedback?
96
00:06:42,940 --> 00:06:48,340
And surprisingly, AI did a pretty good role there, and it helped accelerate quite a bit of her stuff.
97
00:06:49,400 --> 00:06:56,460
Since going into it, I did not know a lot about, let's say, how load cells work or just the tech behind it.
98
00:06:56,520 --> 00:07:01,640
Because when we were reverse engineering stuff, we came across the small parts, which we hadn't really seen before.
99
00:07:02,320 --> 00:07:07,520
So I think doing the initial research and getting some other information and then helping with some initial setup,
100
00:07:07,680 --> 00:07:11,220
because our main rule was to focus more on the device itself.
101
00:07:11,850 --> 00:07:17,000
And the secondary things that came up with it were, let's say, the UI or whatever else.
102
00:07:17,480 --> 00:07:19,860
and so we want to completely focus on the hardware
103
00:07:20,540 --> 00:07:24,840
and AI help in finishing up the secondary task here really quickly
104
00:07:25,020 --> 00:07:26,800
and we were able to write it really quickly
105
00:07:27,320 --> 00:07:30,500
we got the PI to the lab and we were just helping him go through the
106
00:07:30,880 --> 00:07:34,240
the interface and then we got some feedback here that this can be here
107
00:07:34,240 --> 00:07:36,940
this can be done this that was pretty exciting yeah
108
00:07:37,740 --> 00:07:41,620
sort of the AI help with writing the actual circuit Python code that you used on the device
109
00:07:42,780 --> 00:07:47,000
it was a mix honestly um so initially it was just going to
110
00:07:47,060 --> 00:07:49,980
all the different documentation that we could find online for.
111
00:07:50,280 --> 00:07:54,400
So let's say our sensorware at S-H-X-7-11.
112
00:07:55,260 --> 00:07:57,180
So figuring out information for that.
113
00:07:57,500 --> 00:08:03,180
So the initial boilerplate core was provided by AI.
114
00:08:03,400 --> 00:08:04,160
That was pretty helpful.
115
00:08:04,380 --> 00:08:09,860
And
then just fixing it by bed by bit, piece by piece was more of our own experience
116
00:08:10,580 --> 00:08:12,680
and just our own software experience.
117
00:08:13,800 --> 00:08:16,360
You mentioned a couple minutes ago that you had to reverse engineer
118
00:08:16,380 --> 00:08:18,760
the proprietary device, how did that process go?
119
00:08:19,840 --> 00:08:20,620
That was interesting.
120
00:08:20,930 --> 00:08:26,680
So we had, the device, it was like this, imagine a square Mac mini style,
121
00:08:27,060 --> 00:08:29,040
but small Mac mini style device.
122
00:08:29,900 --> 00:08:33,400
And so it was just a lot of unscrewing stuff, taking it all apart.
123
00:08:33,539 --> 00:08:38,039
And then when we opened it, there were these small, small cables inside and these small,
124
00:08:38,240 --> 00:08:38,960
tiny load cells.
125
00:08:39,120 --> 00:08:41,140
And we were like, okay, that seems interesting.
126
00:08:41,900 --> 00:08:45,900
Then the next part was, we had a balance between buying stuff.
127
00:08:46,160 --> 00:08:49,120
that's already out there and making users that and making our own custom things.
128
00:08:49,320 --> 00:08:51,980
Because the whole point of open source was that if we're going to build something
129
00:08:52,100 --> 00:08:53,340
from that, we want it to be accessible.
130
00:08:53,620 --> 00:08:55,800
For example, let's say if we were to go ahead and build our own load cell,
131
00:08:56,540 --> 00:08:59,200
then for someone to build it and replicate it, it's going to be really hard.
132
00:08:59,880 --> 00:09:03,900
So we decide we go on a YouTube, we find a relevant load cell,
133
00:09:04,400 --> 00:09:05,480
and then we'll design it on that.
134
00:09:05,760 --> 00:09:09,100
So that was another key thing that we had to keep in mind.
135
00:09:09,880 --> 00:09:13,020
Then, well, as you were exploring, we're trying to see what materials were used
136
00:09:13,140 --> 00:09:16,100
because the filament that's used to poke on the skin needs to be flexible.
137
00:09:16,120 --> 00:09:20,900
enough, but it also needs to be rigid. So we tried experimenting with different materials and then
138
00:09:21,060 --> 00:09:29,740
eventually, I think we resulted with an aluminum kind of thing. Yeah, to add on to that, I think one of
139
00:09:29,740 --> 00:09:35,560
the challenges, the overarching challenges to everything we would describe was we didn't know what was
140
00:09:35,780 --> 00:09:41,720
good enough, especially because it was medical quality, right? I think if it was another open source
141
00:09:41,920 --> 00:09:46,080
device for tinkering or a watch, I think you can get away with a lot of loose.
142
00:09:46,100 --> 00:09:50,220
or things and can get away with different parts. But we were always like, is this good enough?
143
00:09:50,360 --> 00:09:55,160
Is this accurate enough for medical application? And the Mac mini device that came to us,
144
00:09:55,520 --> 00:09:59,400
as Roque mentioned, like we
had started hacking it apart because we didn't know how it worked.
145
00:09:59,520 --> 00:10:04,520
Like we literally took a saw and cut out pieces of the aluminum and then we like pulled off
146
00:10:04,650 --> 00:10:09,660
the prongs and we started using it ourselves. But it was broken. So we truly had no idea how
147
00:10:09,660 --> 00:10:16,060
it was measuring and how accurate it was. And so when we were building our stuff, we almost hit
148
00:10:16,080 --> 00:10:21,320
a two-month delay or like stock point because we were just not only sure where to proceed.
149
00:10:21,520 --> 00:10:24,580
We were like, we don't want to build our own custom load cell.
150
00:10:25,100 --> 00:10:29,740
You know, we want to use what's off the shelf, but we don't know if this load cell is good enough.
151
00:10:30,320 --> 00:10:37,200
For example, our load cell, I think, in a range of up to, I think, 10K, 1KG or 10KG, if I'm not mistaken,
152
00:10:37,900 --> 00:10:39,700
but we only needed a 10th of that.
153
00:10:40,080 --> 00:10:43,860
We literally only needed 1 to 100 grams of measurement force.
154
00:10:44,560 --> 00:10:49,000
And so we're like, if we're only using such a minimal part of its range, is there too much noise?
155
00:10:49,160 --> 00:10:50,280
Are we able to detect it?
156
00:10:50,760 --> 00:10:52,980
And so we truly had no clue how to reference it.
157
00:10:53,560 --> 00:10:59,260
And then another issue we face with reverse engineering from Blackbox, we thought the whole
158
00:10:59,380 --> 00:11:05,220
time that it was measured in this measurement called like grams force over area to get pressure,
159
00:11:05,380 --> 00:11:06,060
if that makes sense.
160
00:11:06,480 --> 00:11:10,560
So, you know, the tip of the wire is like a diameter of like one millimeter.
161
00:11:11,080 --> 00:11:14,520
And so we're like, oh, we have to calculate the pressure and then calculate.
162
00:11:14,560 --> 00:11:18,020
like the force and divided by that area to get pressure or whatever.
163
00:11:18,460 --> 00:11:22,040
And so we're going through this complicated, convoluted conversion.
164
00:11:23,000 --> 00:11:26,440
And then the clinician comes back and he pulls out the real functioning device
165
00:11:26,800 --> 00:11:28,260
and realized they didn't do any of that.
166
00:11:28,360 --> 00:11:31,600
We realized they were literally just converting directly to force and just doing it
167
00:11:32,290 --> 00:11:33,440
and just reading out that measurement.
168
00:11:34,080 --> 00:11:38,480
And we actually felt totally lied to because all the papers had said that it was over area.
169
00:11:38,840 --> 00:11:41,460
And I was like, wow, no one actually even knows how this device is working.
170
00:11:42,360 --> 00:11:44,180
And they just had some arbitrary readout.
171
00:11:44,580 --> 00:11:50,160
And we found out that our device was almost just as accurate, if not more accurate,
172
00:11:50,620 --> 00:11:53,220
than their current device and was
able to read forces better.
173
00:11:53,860 --> 00:11:56,440
It was able to read closer to zero as well.
174
00:11:57,360 --> 00:12:02,000
And so I think that was one of the big challenges we faced was just validating against
175
00:12:02,280 --> 00:12:04,640
medical accuracy and having medical standards.
176
00:12:05,360 --> 00:12:09,960
And this is important because, you know, as we're trying to build a framework for all
177
00:12:10,020 --> 00:12:14,160
of the open source devices out there, especially medical devices, I think this is a challenge
178
00:12:14,180 --> 00:12:18,560
everyone will face. How can we validate with the scrappiness of open source but also make it
179
00:12:18,700 --> 00:12:23,740
reliable in protection grade for the medical industry. Yeah. And I think like something else that
180
00:12:24,040 --> 00:12:29,500
was while we were going through this process, we, and as we did this, as they called,
181
00:12:29,780 --> 00:12:34,140
contextual inquiry, we were looking at the PI, do we use the device proper on his own. And then
182
00:12:35,120 --> 00:12:40,480
I think there was a part where you had to ask important questions like when it comes to
183
00:12:40,500 --> 00:12:41,300
to accuracy, right?
184
00:12:41,820 --> 00:12:48,200
Something that we realized was, because the device is measuring the sensation over a period of time,
185
00:12:48,760 --> 00:12:51,020
we wanted to know what our tolerance was for accuracy.
186
00:12:51,220 --> 00:12:54,960
Because at the end of the day, fundamentally we're seeing whether the sensation is improving.
187
00:12:55,800 --> 00:13:00,700
So even if, let's say, they were, let's say, a small, I mean, a one person, two person error,
188
00:13:01,260 --> 00:13:04,280
it would have been sort of acceptable because in this particular situation,
189
00:13:05,079 --> 00:13:09,360
the main fundamental use for the device is to see, like, an overall growth over time.
190
00:13:09,960 --> 00:13:14,860
So I think the important part was asking important questions like, hey, what's the real use for it?
191
00:13:14,920 --> 00:13:18,460
That's because it needs to be done this way doesn't really mean that it has to be done that way.
192
00:13:18,900 --> 00:13:25,100
So I think when it comes to open source medical hardware, it's important to understand what your tolerances are, acceptance is are.
193
00:13:25,940 --> 00:13:28,300
And the second part was these guys are really busy.
194
00:13:29,379 --> 00:13:37,680
So we had to really plan one of our first calls with the surgeon was he literally stepped out of the OR just to talk to us for like five minutes.
195
00:13:37,820 --> 00:13:46,400
And it was, it was, it was so I think the second point was, we'd have to plan it out properly, and it was interesting.
196
00:13:48,220 --> 00:13:49,740
So you've been talking a lot about open source.
197
00:13:49,980 --> 00:13:53,460
Will you be certifying the open PSD with the Open Source Hardware Association?
198
00:13:54,900 --> 00:13:58,480
So we did talk to our point of contact with OSHWA.
199
00:13:59,540 --> 00:14:03,400
And he did say that I think OSHWA would be really, really happy to have his certify.
200
00:14:04,020 --> 00:14:07,660
They would also want to do some sort of blog about us and, you know, just,
201
00:14:07,720 --> 00:14:10,420
sort of help us reach out to as many people as possible.
202
00:14:11,000 --> 00:14:15,440
Because the whole point of the NSF grant that was given to OSHWA was to promote such devices.
203
00:14:16,080 --> 00:14:20,900
So, I mean, they said that we're really excited and really happy to collaborate and certify this.
204
00:14:22,060 --> 00:14:24,940
Yeah, I think I remember before we started this project, actually,
205
00:14:25,460 --> 00:14:28,000
we had done an interview with a couple people from OSHWA.
206
00:14:28,520 --> 00:14:33,480
And it was essentially us giving feedback on what we think should be a more certifiable framework
207
00:14:33,520 --> 00:14:37,040
on what open-sourced hardware or medical devices look like.
208
00:14:37,520 --> 00:14:44,100
And it was this ongoing debate because, you know, as we're saying, I think medical devices need more reliability and validation.
209
00:14:45,260 --> 00:14:48,940
But you can only get that validation once you've completed the device, right?
210
00:14:49,300 --> 00:14:57,720
But then part of the ethos of open source is, hey, maybe you get 75% there, you put it online, and then someone else takes it forward.
211
00:14:58,840 --> 00:15:05,840
And so we were having this giant debate as to is validation one of the things that require in order to be certified?
212
00:15:06,660 --> 00:15:17,300
You know, like how does that kind of start veering into, you know, like a research paper publication style, is that too high of a barrier to entry to get people to do a medical device?
213
00:15:18,240 --> 00:15:25,060
And so we're just having this giant debate into what a proper framework for an open source medical device looks like.
214
00:15:25,440 --> 00:15:33,740
So, you know, open to end in a question for the audience to think about, for everyone to think about, what do you really need to certify a reliable open source medical device?
215
00:15:34,420 --> 00:15:36,360
Yeah, for medical, you would want some kind of.
216
00:15:36,380 --> 00:15:39,980
a validation because if you're putting out the design out there and someone builds it and it
217
00:15:40,120 --> 00:15:45,100
doesn't do what it's supposed to do, even though it's an open source license and it could be built,
218
00:15:45,819 --> 00:15:50,600
they're still not getting what they expect. Exactly. And we, I think, had some other problems
219
00:15:50,820 --> 00:15:55,880
too. We are saying when we're building this wire prong, originally we were thinking of creating
220
00:15:56,100 --> 00:16:01,500
this giant wire bending jig when we're going to screw some things in and then it would wire
221
00:16:01,760 --> 00:16:06,340
automatically with a feed machine. But then we were thinking and we're like, well, how many people
222
00:16:06,360 --> 00:16:10,780
we're going to have this complex wire bending jig that the resources of Cornell do.
223
00:16:11,360 --> 00:16:14,240
And so we ended up just buying jewelry pliers on Amazon.
224
00:16:14,500 --> 00:16:16,980
And we're like, okay, well, just hand wire it, hand bend it.
225
00:16:17,220 --> 00:16:21,840
But then in that process, we're like, how can we communicate across the screen that this is
226
00:16:21,980 --> 00:16:23,900
exactly the dimensions that they need to do it at?
227
00:16:24,320 --> 00:16:28,780
And so I literally ended up tracing the shape of the wire on a piece of paper and then putting
228
00:16:28,880 --> 00:16:30,800
like a quarter next to it, took a photo.
229
00:16:30,900 --> 00:16:34,960
And I was like, okay, everyone, print this out, you know, zoom into your photo to size.
230
00:16:35,120 --> 00:16:36,560
and this is how you're going to bend your wire.
231
00:16:37,440 --> 00:16:39,640
And so it was trying to assess the accuracy,
232
00:16:39,940 --> 00:16:42,700
but also balance the scrappiness of what open source is.
233
00:16:45,160 --> 00:16:48,800
Key part here is the way we convinced the surgeons
234
00:16:49,100 --> 00:16:50,540
and the doctors to come work with us
235
00:16:50,620 --> 00:16:51,980
because we wanted to keep them,
236
00:16:52,030 --> 00:16:55,160
I mean, place them on the driving seat
237
00:16:55,340 --> 00:16:57,660
and draw the device around their use.
238
00:16:57,820 --> 00:16:58,920
I think that was really important.
239
00:16:59,060 --> 00:17:01,200
A lot of times medical equipment that would be designed
240
00:17:01,930 --> 00:17:04,920
are really far apart from what the actual use case.
241
00:17:05,020 --> 00:17:06,740
sometimes. I mean, that's something interesting.
242
00:17:07,120 --> 00:17:08,060
Yeah, we can focus on it.
243
00:17:09,420 --> 00:17:11,260
Michelle, any thoughts as we wrap up?
244
00:17:12,140 --> 00:17:13,939
There were other things that we also had to do,
245
00:17:14,120 --> 00:17:17,860
including the 3-D printing, the form factor prototyping as well.
246
00:17:18,500 --> 00:17:21,100
I think one thing kind of, I think,
247
00:17:21,400 --> 00:17:25,100
elaborating into that tension of scrappiness versus medical grade hardware,
248
00:17:26,480 --> 00:17:29,500
you know, we were literally just building in the maker lab.
249
00:17:29,720 --> 00:17:33,480
Like, we're quite literally, you know, just scrapping it up in the maker lab.
250
00:17:34,260 --> 00:17:40,200
And so all of our prototypes for the physical, like the hand-held portion and casing the hardware.
251
00:17:40,760 --> 00:17:41,860
It was all 3D-printed.
252
00:17:41,980 --> 00:17:45,060
And honestly, it had like a lot of bumps and I tried to sand it out.
253
00:17:45,200 --> 00:17:48,960
But it just kept making it worse and worse because the dust is getting
everywhere and into the crevices.
254
00:17:50,120 --> 00:17:56,780
And so highlighting that tension, again, we're like, how will a patient feel if their doctor's holding a little 3D-printed screwdriver toy?
255
00:17:57,560 --> 00:18:00,100
And then we're like, how will this hold up over the years?
256
00:18:00,380 --> 00:18:03,980
And that was one of the issues and the tensions we had faced as well.
257
00:18:04,200 --> 00:18:33,980
So I think as a clinician right now is using it, we're excited to get feedback. And perhaps we'll have more resourcing to cast it better or, you know, wrap it up in silicone. But I think our first version was definitely perhaps not built to last, to say the least. And I think adding on to that tension as well, me and Rick are both in grad school and this is our final year here. And so we are wondering about the open source maintenance of a medical device as well. And we were wondering,
258
00:18:34,020 --> 00:18:39,900
like who will carry this forward. You know, the clinician and, you know, they're not necessarily
259
00:18:39,990 --> 00:18:44,260
the most technical people in terms of being able to install the software, the software breaks.
260
00:18:45,440 --> 00:18:49,720
And so one of the kind of lasting questions in the back of our head is how is this going to go
261
00:18:50,930 --> 00:18:54,580
forward? And that was originally one of the issues, the original proprietary medical device,
262
00:18:54,900 --> 00:18:58,120
was that they're being charged every single year, essentially, to have some sort of maintenance
263
00:18:58,480 --> 00:19:03,980
certification. But we realized the certification didn't actually do anything. Like the device is so simple
264
00:19:04,000 --> 00:19:08,880
of load cell and the calibration, all you have to do in order to calibrate it is hang away on
265
00:19:08,960 --> 00:19:12,980
it. And that will be your reference point. And once you reference point it, it'll just reset the
266
00:19:14,300 --> 00:19:17,820
calibration. It was so simple. I was like, no one needs to be charged for this. This device is none
267
00:19:17,940 --> 00:19:23,320
we sent in to maintain every single year. But in the back of my mind is what are we going to do?
268
00:19:23,500 --> 00:19:27,240
How are we going to communicate the disclination? How are we going to validate that there has been
269
00:19:27,740 --> 00:19:32,919
no measurement drift across time? And so that has been one of our longstanding questions of open
270
00:19:32,980 --> 00:19:35,120
source as well, especially in the medical use case.
271
00:19:35,820 --> 00:19:36,640
Yeah, that's a great question.
272
00:19:36,760 --> 00:19:39,340
I know I don't have any answers off the top of my head for that.
273
00:19:39,900 --> 00:19:44,200
At least talking about the wear and tech, one good thing about open sourcing
274
00:19:44,900 --> 00:19:47,540
this is even if it wears down, you can just reprint it.
275
00:19:47,820 --> 00:19:49,220
That's like the part about this.
276
00:19:49,660 --> 00:19:51,580
Like the main core part still stays.
277
00:19:52,040 --> 00:19:55,720
They can just reprint the whole outer casing if they want to with different colors.
278
00:19:56,340 --> 00:20:00,559
Honestly, if they want to use it on different age groups of patients, they can make it
279
00:20:00,560 --> 00:20:01,980
if they want to do it.
280
00:20:02,070 --> 00:20:09,080
So there's this endless possibilities in terms of making the device look approachable and safe.
281
00:20:09,820 --> 00:20:11,500
What's next for you with the device?
282
00:20:12,940 --> 00:20:17,160
I think what's next right now is, as we mentioned, because it's a medical device,
283
00:20:17,900 --> 00:20:20,360
we really invested into making really extensive documentation.
284
00:20:20,880 --> 00:20:24,500
As I mentioned, even drawing out how the wire should be bent.
285
00:20:24,980 --> 00:20:27,900
For example, we had to drill a hole into our load cell.
286
00:20:28,600 --> 00:20:32,100
And I was like, how do I map out and communicate to someone where to drill this hole
287
00:20:32,180 --> 00:20:35,820
and like how deep the hole should be and all of these things?
288
00:20:36,500 --> 00:20:39,560
And so we really invested a lot of energy into making extensive documentation.
289
00:20:40,360 --> 00:20:42,600
So we're hoping to get like a V1, V2 out soon.
290
00:20:43,260 --> 00:20:46,480
Other things that are next is obviously iterating clinician feedback.
291
00:20:46,980 --> 00:20:49,720
Other things is also building that framework out with OSHWA.
292
00:20:50,220 --> 00:20:51,680
So scaling the impact.
293
00:20:52,320 --> 00:20:56,340
And we're also interested in perhaps some sort of publishing commentary that we're working on.
294
00:20:57,660 --> 00:21:05,100
to I guess propose this new framework as we're saying of how open source has enabled participatory med tech development.
295
00:21:05,840 --> 00:21:15,700
In contrast to what industry proprietary black box looks like, it's you build the device and you put it and you sell it to hospitals that are super expensive.
296
00:21:16,380 --> 00:21:20,820
I'm highly dubious that there's any sort of clinician iteration loop that's happening there.
297
00:21:21,420 --> 00:21:24,340
And so we really wanted to publish on what a new framework could look like.
298
00:21:25,140 --> 00:21:27,520
Yeah, and we're also waiting to hear back in the BI.
299
00:21:27,540 --> 00:21:34,020
because I think he was planning on using this as part of his research as well as he uses it on patients.
300
00:21:34,720 --> 00:21:38,640
So we're looking forward to his findings and his results and how that goes.
301
00:21:39,240 --> 00:21:40,020
Yeah, super exciting.
302
00:21:40,960 --> 00:21:41,600
That is exciting.
303
00:21:42,320 --> 00:21:44,320
Reitwiec and Michelle, thanks so much for coming on the show.
304
00:21:45,320 --> 00:21:46,720
Yeah, thank you so much for having us.
305
00:21:47,100 --> 00:21:48,760
That was a really fun chatting about all this.
306
00:21:50,960 --> 00:21:52,920
Thank you for listening to the Circuit Python Show.
307
00:21:53,420 --> 00:21:55,680
For show notes and transcript, visit www.
308
00:21:55,720 --> 00:21:57,660
www.Cythonshow.com.
309
00:21:58,200 --> 00:21:59,680
Until next time, stay
310
00:22:25,020 --> 00:22:25,280
positive.