رونوشت 🔗
WEBVTT
00:00:00.959 --> 00:00:03.200
SwissInfo podcasts.
00:00:03.680 --> 00:00:11.199
According to the experts I spoke to, AI will not take doctors' jobs, but help them get better, faster, and more accurate results.
00:00:12.720 --> 00:00:17.519
When someone has cancer, any delay in diagnosis can reduce the chances of recovery.
00:00:17.679 --> 00:00:19.920
This is where artificial intelligence can help.
00:00:20.079 --> 00:00:24.480
AI software can detect patterns in medical images that the human eye may miss.
00:00:31.760 --> 00:00:33.039
So you're taking a risk.
00:00:33.439 --> 00:00:35.920
Either nothing is found or we find something.
00:00:36.079 --> 00:00:37.200
But it's a rare condition.
00:00:37.280 --> 00:00:39.520
And we don't really know how to treat rare conditions.
00:00:40.399 --> 00:00:42.479
Look at how different we are on the outside.
00:00:42.560 --> 00:00:45.840
And like, what do you think, therefore, we look like on the inside?
00:00:46.000 --> 00:00:58.960
Our science has advanced enough so we can find abnormal cells at very early stages, but it hasn't advanced enough to distinguish which early cells will grow quickly and threaten life.
00:01:09.920 --> 00:01:13.599
Artificial intelligence is a big game changer for doctors.
00:01:13.760 --> 00:01:19.280
Thanks to AI, diagnosing illnesses can be done much faster and more accurately.
00:01:19.439 --> 00:01:22.879
And illnesses can be detected at an earlier stage too.
00:01:23.120 --> 00:01:24.640
But how does it work?
00:01:24.879 --> 00:01:27.040
And is this really good for us?
00:01:36.879 --> 00:01:41.760
Hello, I'm Joe Fay, and you're listening to the Swiss Connection Science Podcast.
00:01:42.079 --> 00:01:47.439
Nearly half of the global population doesn't have access to appropriate diagnostics.
00:01:47.599 --> 00:01:52.319
That's because the field is underfunded, and there's a lack of pathology experts.
00:01:52.640 --> 00:01:58.879
Delays in diagnosis can push back life-saving care and costs health systems millions of dollars.
00:01:59.120 --> 00:02:01.680
But this could soon change thanks to AI.
00:02:01.920 --> 00:02:09.599
For instance, stethoscopes can now hear heartbeats that humans can't, or match lung sounds with those of respiratory diseases.
00:02:09.840 --> 00:02:14.800
My colleague Eileen Elche has been looking into this topic, and she's now in the studio with us.
00:02:14.960 --> 00:02:15.680
Welcome, Eileen.
00:02:16.159 --> 00:02:16.800
Hi, Joe.
00:02:16.879 --> 00:02:17.599
It's nice to be here.
00:02:18.000 --> 00:02:21.120
So tell us, Eileen, where do we stand with the technology today?
00:02:21.360 --> 00:02:24.400
Are we going to see job cuts for doctors because of this?
00:02:24.800 --> 00:02:27.680
So, according to the experts I spoke to, no.
00:02:28.080 --> 00:02:34.080
AI will not take doctors' jobs, but help them get better, faster, and more accurate results.
00:02:34.400 --> 00:02:44.000
I recently visited Geneva Hospital and asked the head
Load full transcript
of the pathology division, Professor Laura Rubia-Brandt, how AI has changed their work.
00:02:50.080 --> 00:02:53.439
A diagnosis must be both fast and accurate.
00:02:53.680 --> 00:02:58.560
When someone has cancer, any delay in diagnosis can reduce the chances of recovery.
00:02:59.039 --> 00:03:06.719
That's why technology and medical innovation have played just as important a role in advancing health care as the discovery of new drugs.
00:03:06.960 --> 00:03:09.520
Together, they have helped increase life expectancy.
00:03:13.120 --> 00:03:14.159
Let me give you an example.
00:03:14.719 --> 00:03:17.199
One in eight women will develop breast cancer.
00:03:17.280 --> 00:03:18.639
That's a huge number.
00:03:18.879 --> 00:03:20.000
One in eight.
00:03:20.240 --> 00:03:24.159
And unfortunately, we are seeing it affect younger and younger women.
00:03:24.479 --> 00:03:31.439
Cancers that were once mainly diagnosed after the age of 50 are now appearing more often in younger pre-menopausal women.
00:03:31.919 --> 00:03:34.879
So what is the best chance we can give these women?
00:03:35.120 --> 00:03:37.919
It's not to screen everyone with mammograms.
00:03:38.159 --> 00:03:43.360
In theory, we could, and we would detect very small lesions long before they can be felt.
00:03:43.599 --> 00:03:45.919
But in practice, that isn't realistic.
00:03:46.080 --> 00:03:47.599
The costs are too high.
00:03:47.759 --> 00:03:53.840
It requires too many resources, and repeated screening would expose patients to unnecessary radiation.
00:03:54.319 --> 00:03:58.080
Instead, we need to identify the women who are most at risk.
00:03:58.319 --> 00:04:02.479
Out of 100 women, which ones are more likely to develop cancer?
00:04:03.039 --> 00:04:05.439
This is where artificial intelligence can help.
00:04:05.680 --> 00:04:10.639
AI software can detect patterns in medical images that the human eye may miss.
00:04:10.879 --> 00:04:14.080
It can spot subtle changes that are difficult for me to see.
00:04:14.319 --> 00:04:21.199
When I examine a mammogram, the software might tell me there's something here that suggests this patient could be at higher risk.
00:04:21.360 --> 00:04:23.040
And that's the patient I need to focus on.
00:04:27.680 --> 00:04:36.160
Geneva Hospital's pathology division was the first in Switzerland and one of the few in Europe to fully digitalize its workflow five years ago.
00:04:38.399 --> 00:04:40.160
So are you already using AI today?
00:04:41.040 --> 00:04:42.160
I mean on the clinical side.
00:04:44.079 --> 00:04:45.199
Yes, every day.
00:04:45.360 --> 00:04:49.519
We use it for everything related to pre-screening and workflow management.
00:04:49.759 --> 00:04:56.959
Anything that helps us save time, detecting cell proliferation, identifying bacteria, or flagging blurry slides.
00:04:57.279 --> 00:04:57.920
Absolutely.
00:04:58.000 --> 00:04:58.959
It's part of our daily work.
00:05:03.680 --> 00:05:08.560
Rubia Brandt keeps her microscope in a
corner of her office in case of emergency.
00:05:08.720 --> 00:05:11.839
But she says it's mostly been gathering dust.
00:05:12.079 --> 00:05:15.199
We walked over to her computer to look at some cases.
00:05:27.279 --> 00:05:29.519
Oh, the number of images is enormous.
00:05:29.759 --> 00:05:33.439
In terms of cases, though, it's usually between five and twenty-five a day.
00:05:34.879 --> 00:05:40.560
Look, for example, here we can see a photograph of the tissue that has been removed from a patient during an operation.
00:05:40.800 --> 00:05:42.639
In this case, it's a piece of the liver.
00:05:43.279 --> 00:05:45.439
And here are the samples under the microscope.
00:05:45.920 --> 00:05:48.319
So what you see here are cells.
00:05:53.759 --> 00:05:54.800
There are quite a lot of them.
00:05:55.199 --> 00:05:55.839
Oh yes.
00:05:56.000 --> 00:05:56.959
It's a big liver.
00:05:57.279 --> 00:05:58.720
Here, I'm looking at the cells.
00:05:58.959 --> 00:06:04.399
Then I might say to myself, okay, I need to know the distance between these two points.
00:06:04.639 --> 00:06:07.279
Here it's 0.4 millimeters.
00:06:07.519 --> 00:06:09.279
Or I might need to count cells.
00:06:10.160 --> 00:06:12.720
Here perhaps 2,000 cells.
00:06:12.959 --> 00:06:14.879
Over there, 1,000.
00:06:15.120 --> 00:06:18.639
And in this little circle, there are about 100 cells.
00:06:18.959 --> 00:06:24.959
Elsewhere, I might need to count 200, 500, or 1,000 cells.
00:06:25.120 --> 00:06:27.920
So I click here and the software does it for me.
00:06:28.240 --> 00:06:29.920
Analysis in progress.
00:06:30.160 --> 00:06:30.959
And there we go.
00:06:31.279 --> 00:06:36.079
It counted 500 cells, and 0.2% of them are currently proliferating.
00:06:43.600 --> 00:06:53.759
If I had to sit here and count all of those manually, counting 1,000 cells or taking these kinds of measurements is extremely difficult and very time consuming.
00:06:55.199 --> 00:06:58.079
And five years ago, were you still doing all of this manually?
00:06:58.399 --> 00:06:58.879
Absolutely.
00:06:59.439 --> 00:07:05.680
For example, let's say you have a melanoma and I need to determine how deeply it has invaded the tissue.
00:07:05.920 --> 00:07:09.279
One millimeter, two millimeters, three millimeters.
00:07:09.519 --> 00:07:13.439
The deeper the tumor goes, the more it can affect the treatment plan.
00:07:13.600 --> 00:07:20.959
If the tumor is very deep, even with clear margins, I may need to remove nearby lymph nodes to check whether the cancer has spread.
00:07:21.199 --> 00:07:25.439
In the past, I would literally use a ruler and a pencil to measure everything.
00:07:25.600 --> 00:07:26.560
Can you imagine?
00:07:26.720 --> 00:07:28.399
Down to the millimeter.
00:07:28.800 --> 00:07:31.839
Now, this particular image isn't skin tissue.
00:07:32.000 --> 00:07:38.000
But here, for example, I can simply tell the software, measure the distance between this point and that point.
00:07:38.079 --> 00:07:39.839
And it calculates it instantly.
00:07:40.240 --> 00:07:41.600
1.3 centimeters.
00:07:46.399 --> 00:07:53.040
So when you say you review between 5 and 25 cases a day, roughly how many images does that represent per case?
00:07:54.079 --> 00:07:57.519
Typically anywhere from 5 to 50 images per case.
00:08:23.920 --> 00:08:30.800
Where we still have a big question mark is whether AI will one day provide diagnostic insights that we simply can't see ourselves.
00:08:30.959 --> 00:08:32.960
That's still very much a work in progress.
00:08:37.279 --> 00:08:41.679
For example, predicting whether a particular person is more likely to develop cancer?
00:08:42.240 --> 00:08:42.879
Exactly.
00:08:43.120 --> 00:08:45.039
Right now, we don't know yet.
00:08:45.200 --> 00:08:48.240
We're only just beginning to explore those possibilities.
00:08:48.399 --> 00:08:51.919
But in five or ten years, it may well become a reality.
00:08:52.240 --> 00:08:56.720
We still need to conduct studies and we need to test these technologies thoroughly.
00:08:56.960 --> 00:09:02.559
Most importantly, AI tools need to be adapted to our own population and healthcare environment.
00:09:02.799 --> 00:09:11.120
Many AI systems are developed using American patient and healthcare data, for example, and that can be quite different from what we see here.
00:09:11.440 --> 00:09:13.200
A simple example is body weight.
00:09:13.360 --> 00:09:17.279
Obesity rates are much higher in the United States than they are here.
00:09:17.519 --> 00:09:20.559
The demographic makeup of the population is also different.
00:09:20.799 --> 00:09:25.360
We need to understand how all of these factors influence the performance of these systems.
00:09:25.519 --> 00:09:32.799
That's why it's so important for us to develop expertise ourselves, rather than relying entirely on solutions developed elsewhere.
00:09:33.039 --> 00:09:39.759
And what does that mean for someone like me, a doctor who's not a computer scientist and doesn't know how to code?
00:09:40.080 --> 00:09:43.039
It means bringing those skills directly into the hospital.
00:09:43.759 --> 00:09:52.159
For example, in my department, I've created positions for physician engineers and integrated software engineers directly into our medical teams.
00:09:52.480 --> 00:09:55.360
These people work side by side with clinicians in the field.
00:09:58.960 --> 00:10:03.519
And ideally, that allows them to develop software based on data coming directly from Switzerland.
00:10:04.159 --> 00:10:04.320
Exactly.
00:10:04.639 --> 00:10:04.879
Exactly.
00:10:05.519 --> 00:10:06.879
And that's becoming more and more important.
00:10:08.559 --> 00:10:11.759
It's really amazing what technology can do nowadays, isn't it?
00:10:11.919 --> 00:10:18.000
And this really seems to speed up things quite a lot, and it'll hopefully help improve people's health overall as well.
00:10:18.159 --> 00:10:20.399
So we're going to change the focus a bit here.
00:10:20.639 --> 00:10:31.120
New AI capabilities also boost older inventions, like MRI scanners, and they're all linked to the ongoing trend of people wanting to live longer and healthier lives.
00:10:31.360 --> 00:10:35.440
Health clinics are offering full-body scans to totally healthy people.
00:10:35.600 --> 00:10:41.600
And the key phrase here is early detection, which sometimes can do more harm than good.
00:10:42.240 --> 00:10:42.720
Right.
00:10:42.879 --> 00:10:48.480
These scans are usually offered by private companies or startups and cost about$1,000.
00:10:49.200 --> 00:10:50.240
I spoke to Dr.
00:10:50.320 --> 00:10:52.080
Francis Mayer about this.
00:10:52.320 --> 00:10:59.039
He's the co-founder of the Swiss Center for Preventive Medicine at the Hôpital de la Tour Campus in Geneva.
00:10:59.360 --> 00:11:07.679
He and his colleagues put a lot of thought into when early detection really serves a patient, but decided it is often just good for business.
00:11:16.480 --> 00:11:18.799
In women, breast cancer is one of the most common.
00:11:19.360 --> 00:11:21.120
In men, prostate cancer.
00:11:21.279 --> 00:11:25.279
In both men and women, we routinely screen for colorectal cancer.
00:11:25.519 --> 00:11:30.639
Lung cancer, particularly in smokers, and skin cancer are also major concerns.
00:11:30.799 --> 00:11:35.200
For these common cancers, established screening programs already exist.
00:11:35.440 --> 00:11:42.080
In some cantons, such as Geneva, people receive invitations for screening once they reach a certain age.
00:11:42.399 --> 00:11:52.240
Women may be invited for mammography screening, while both men and women may be offered colorectal cancer screening, either through a colonoscopy or a stool test.
00:11:52.399 --> 00:11:57.039
Now, when you undergo a whole-body CT scan, there are two possible outcomes.
00:11:57.279 --> 00:11:59.519
The first is that the scan finds nothing at all.
00:11:59.679 --> 00:12:01.279
That's what everyone hopes for.
00:12:01.519 --> 00:12:04.240
The only downside is that you've been exposed to radiation.
00:12:09.519 --> 00:12:14.960
There are concerns that some cancers may be induced by medical radiation exposure.
00:12:15.200 --> 00:12:22.879
MRIs or magnetic resonance imaging doesn't use ionizing radiation, making it safer for repeated use.
00:12:23.360 --> 00:12:29.919
But MRIs often still require the use of contrast agents to make certain abnormalities more visible.
00:12:30.240 --> 00:12:37.519
Mayer points out that some people can also have allergic reactions to these agents, including potentially severe reactions.
00:12:39.440 --> 00:12:40.879
So you're taking a risk.
00:12:41.120 --> 00:12:43.519
Either nothing is found or something is found.
00:12:43.679 --> 00:12:48.320
If something is found, then we have to investigate further, because a lump has been detected.
00:12:48.720 --> 00:12:50.399
That usually means a biopsy.
00:12:50.639 --> 00:12:58.240
A biopsy can be painful, can cause bleeding and bruising, and it may
even lead to complications such as infection.
00:12:58.559 --> 00:13:00.639
The biopsy will then tell us two things.
00:13:01.679 --> 00:13:08.480
Either there's nothing there, in which case we're back to square one, except that we've performed an invasive procedure.
00:13:08.720 --> 00:13:20.080
The patient has been punctured, costs have been incurred, the biopsy had to be carried out, often under CT or ultrasound guidance, the sample had to be analyzed, and the laboratory had to be paid.
00:13:21.519 --> 00:13:23.039
All that, only to find nothing.
00:13:24.080 --> 00:13:25.519
Or we find something.
00:13:25.759 --> 00:13:27.279
But it's a rare condition.
00:13:27.519 --> 00:13:31.679
And we don't really know how to treat rare conditions, so we treat them as best we can.
00:13:32.559 --> 00:13:35.039
We may do chemotherapy, radiotherapy.
00:13:35.279 --> 00:13:37.279
We may operate and remove the lump.
00:13:37.519 --> 00:13:39.440
We may try a range of interventions.
00:13:39.600 --> 00:13:44.240
But let's say that person had an identical twin who never underwent a whole body scan.
00:13:44.879 --> 00:13:47.360
Both may ultimately die at exactly the same time.
00:13:48.320 --> 00:13:53.519
The difference is that the first twin becomes a patient the moment the scan or biopsy reveals something abnormal.
00:13:53.679 --> 00:13:58.080
The second twin, who never asks for the scan, remains at peace and feels perfectly well.
00:13:58.320 --> 00:14:06.720
And that period of feeling well could last days, weeks, months, or even years before the cancer or disease we detected would even have caused symptoms.
00:14:10.799 --> 00:14:14.960
These scans have partly been boosted by influencers like Kim Kardashian.
00:14:15.200 --> 00:14:28.879
In Switzerland, where private clinics, hospitals, and luxury wellness centers have been offering preventive health treatments for decades, at least five different startups offering full-body scans have popped up in the last four years.
00:14:29.120 --> 00:14:36.080
I spoke to Susan O'Sullivan, a consultant neurologist at the University College London about the trend.
00:14:36.399 --> 00:14:41.200
What I always say to people is that look at how different we are on the outside.
00:14:41.279 --> 00:14:45.919
I mean, myself and, you know, Kim Cardassian, we could not look more different.
00:14:46.080 --> 00:14:49.360
And like, what do you think, therefore, we look like on the inside?
00:14:49.600 --> 00:14:57.360
You know, all of these little scars and differences, you know, some people have um, you know, some people are one shape, some people are the other shape.
00:14:57.519 --> 00:15:02.320
The inside is equally fraught with kind of individual differences.
00:15:02.799 --> 00:15:06.000
And you have to realize that we didn't even have things like an M.
00:15:06.240 --> 00:15:11.279
So MRI scans only came into regular clinical use in the 1990s.
00:15:11.519 --> 00:15:16.879
Um, but you know, even then in the 1990s, they were quite blunt instruments.
00:15:17.039 --> 00:15:20.480
You know, the images we'd look back at them now and say they were very fuzzy.
00:15:20.720 --> 00:15:29.120
Um, you know, it's really only in the last so say 10 years or so that we've had, you know, these incredibly good images that have been widely available.
00:15:29.360 --> 00:15:43.759
So that's how long we've had to find out what before something like the MRI scan, we didn't even know what the inside of the healthy body looked like because you you couldn't do CT scans on everyone because they came with big doses of radiation.
00:15:44.320 --> 00:16:15.840
So this is we've we are so early in this stage of scanning the inside of people, and therefore, we are only now beginning to learn about all those internal differences, and we are only now learning how many people have lived out their entire lives with cysts and blood vessel abnormalities and spots and dots and um things inside us that we've never seen before, and we are just now learning of them and trying to figure out what they are.
00:16:16.080 --> 00:16:35.039
And if you have a scan and a doctor finds something and they don't know quite what it is or quite how it will progress, it will very often lead to treatment because you know, as a doctor, you don't want to be the person who found something, didn't do anything about it, and then it turned out to be life-threatening.
00:16:35.200 --> 00:16:41.360
So you will always err on the side of seeing these abnormalities and thinking, well, I better treat it just in case.
00:16:41.759 --> 00:17:00.879
Our science has advanced enough so we can find abnormal cells at very early stages, but it hasn't advanced enough to distinguish which early cells will grow quickly and threaten life, and which early cells will just stay more or less the same for the rest of your life and not necessarily need treatment.
00:17:01.279 --> 00:17:08.799
So our assumption that you're catching things early will always help hasn't proven over time to be correct.
00:17:08.960 --> 00:17:21.599
We can see that um we do not know the natural history of every disease, and we do not know how every s every kind of abnormality will progress for each individual.
00:17:22.400 --> 00:17:33.680
Over-diagnosis, the treatment of a disease that won't progress to cause symptoms, stems from a well-meaning intention to detect illness early and save lives.
00:17:34.240 --> 00:17:43.920
But according to O'Sullivan, the assumption that catching things early will always help hasn't proven to be correct over time for every cancer.
00:17:44.400 --> 00:17:48.559
O'Sullivan recently published a book on the topic of over-diagnosis.
00:17:48.880 --> 00:17:53.119
So I think it's really important people understand what over-diagnosis is.
00:17:53.200 --> 00:18:08.160
So I think what people think of when they think of overdiagnosis is they think it means there's either there's nothing wrong with you or there's um you're complaining about
nothing, the diagnosis is wrong, snap out of it, that sort of thing.
00:18:08.319 --> 00:18:09.440
That isn't what it means.
00:18:09.680 --> 00:18:21.200
Overdiagnosis just recognizes the point at which a diagnosis, which may be perfectly right, um, isn't actually helpful and could be detrimental to your health.
00:18:21.519 --> 00:18:27.920
A diagnosis changes your relationship with your body and it changes your relationship with your health.
00:18:28.240 --> 00:18:34.559
So the minute you give someone a diagnosis, you potentially it depends on a person, people will respond differently.
00:18:34.880 --> 00:18:38.480
But when you give someone a diagnosis, you turn them into a patient.
00:18:38.720 --> 00:18:47.599
And so for some people, that risks medicalizing their lives unnecessarily, and some people will start behaving like a patient.
00:18:47.839 --> 00:18:52.960
So one of the risks would be the nocebo effect or the labeling effect or the classification effect.
00:18:53.039 --> 00:19:04.720
So if I tell someone they have a genetic problem that causes difficulty walking, and in time they'll they'll have more difficulty walking, that might reinforce those difficulties by forcing them to pay attention to them.
00:19:05.039 --> 00:19:08.400
Whereas normally you might just not notice something and overcome it.
00:19:09.039 --> 00:19:14.240
Talking about giving it a label can focus on the symptoms and make them worse instead of better.
00:19:14.559 --> 00:19:23.920
The minute I give someone a diagnosis, I'm affecting their travel insurance, their health insurance, um, their application for a mortgage.
00:19:24.079 --> 00:19:30.960
So there's so many, there are both physical downsides in terms of how it changes the way you trust your body.
00:19:31.119 --> 00:19:38.640
And then there's all these practical downsides in terms of um now you're a patient, you have to tell other people you're a patient, and that will impact on you also.
00:19:38.880 --> 00:19:43.359
So we should think if you if you're unwell, you need a diagnosis.
00:19:43.519 --> 00:19:52.480
But if you're in that borderline zone where you don't actually have any symptoms and you're not particularly unwell, we should be thinking about these negative impacts a little bit more.
00:19:52.960 --> 00:19:57.839
How do you draw the line between prevention and over-diagnosis?
00:19:58.000 --> 00:20:03.759
How do you make sure that people still get the help they need or the do the screenings they should do?
00:20:04.079 --> 00:20:04.559
Yeah.
00:20:04.880 --> 00:20:08.000
Yeah, I mean, that honest to God, that worries me too.
00:20:08.160 --> 00:20:13.359
You know, I don't want to write about health screening and scare people away from it.
00:20:13.519 --> 00:20:21.759
Um, so I very much encourage people to have whatever screening their individual kind of country or health service recommends.
00:20:21.920 --> 00:20:25.279
Um, I want to say first, though, that the word prevention is wrong.
00:20:25.440 --> 00:20:31.519
Uh, you know, prevention means, you know, stop smoking, don't drink, exercise, and and so forth.
00:20:31.759 --> 00:20:40.160
Um, what you're talking about is is screening, which isn't prevention, that's diagnosing things uh at a at the earliest possible stage.
00:20:40.400 --> 00:21:15.039
So I think what what I say to people on this, on the subject of, you know, um, how do you decide whether you're going to have the screening tests or not, is that you have what your health service recommends, but you you go into this knowing that there are some uncertainties there, so that you're prepared to ask your doctor the right question, first of all, and secondly, so that you are um kind of armed to think about the test results when you get them, um, and know that you don't have to take immediate action.
00:21:15.119 --> 00:21:20.400
So, if I use the example of breast cancer screening, for example, which is the one that always scares me.
00:21:20.480 --> 00:21:30.319
I mean, I'm a woman, I also don't want to get breast cancer, and I also worry about what I should do for me myself personally with regard to screening.
00:21:30.559 --> 00:21:43.839
So, what what I do personally is I is I agree to the screening that the NHS um recommends to me, which is to, you know, have uh mammography every three years.
00:21:44.000 --> 00:21:59.759
Um, but what I'm prepared to do is if I get that positive result, is um discuss it with my doctor, understanding that a large percentage of women who who have abnormalities on screening do not necessarily.
00:22:00.160 --> 00:22:07.519
Necessarily progress to malignant cancer that grows and affects your health.
00:22:07.680 --> 00:22:14.400
And I'm prepared to ask my doctor for watchful waiting, since that would be my preference.
00:22:14.559 --> 00:22:17.759
So in the case of breast cancer, know yourself.
00:22:17.920 --> 00:22:29.680
If you are someone who really worries about your health and they couldn't live with having an abnormal lump in the breast, you know, waiting to see what happens next, then you go ahead and you have the treatment that is recommended.
00:22:29.920 --> 00:22:41.680
Or if perhaps you're like me and you would prefer to know that you definitely need all of that invasive treatment before you have it, you might ask to enter a watchful waiting program, which means waiting for a while.
00:22:41.839 --> 00:22:52.720
So it's about understanding the issues and knowing how you prefer to manage your own health problems so that you can individualize the decision about what happens next to yourself.
00:22:53.599 --> 00:22:56.960
How would you avoid overdiagnosis?
00:22:57.359 --> 00:23:04.880
Well, I think what we need to do better is to try and figure out, you know, what what disease progresses and which diseases don't.
00:23:05.599 --> 00:23:08.000
At some point we will understand that better.
00:23:08.240 --> 00:23:14.319
But for the moment, I think we would do better to do screening on high-risk individuals.
00:23:14.480 --> 00:23:26.559
So for example, if your blood pressure is a little over the cutoff point, but you are a healthy weight, you don't smoke, you exercise, you have no family history of stroke or heart disease.
00:23:26.720 --> 00:23:35.599
So your only thing is that you've got borderline high blood pressure, then you're not a high-risk individual and you perhaps did not benefit from that diagnosis.
00:23:36.000 --> 00:23:46.799
Where we should be focusing this kind of screening and diagnosis on people who have multiple risk factors and therefore are more likely to be the ones who need treatment.
00:23:47.039 --> 00:23:57.759
Very often, when we look at screening programs and assess their efficacy, we assess how many people they can find rather than how many people benefited from that finding.
00:23:57.920 --> 00:24:03.759
And I think high-risk screening is um it would be a better strategy.
00:24:11.279 --> 00:24:14.559
Assessing the risks could be a better strategy for the future.
00:24:14.720 --> 00:24:17.039
So can AI help us with that?
00:24:17.519 --> 00:24:18.319
Absolutely.
00:24:18.559 --> 00:24:26.799
I think this really loops back to what Professor Rubia Brandt was saying about AI capabilities and predicting how cells will develop.
00:24:27.119 --> 00:24:45.279
The Federal Institute of Technology Zurich, ETH, and the Massachusetts Institute of Technology, MIT, collaborated on a tool that can determine whether early-stage, non-invasive breast cancers will develop into invasive ones thanks to patterns that AI recognizes.
00:24:45.599 --> 00:24:57.279
And the University of Geneva recently developed an AI model called mangrove gene signatures, which is capable of understanding the logic and form of colon cancers.
00:24:57.519 --> 00:24:59.920
Both are still in research phase.
00:25:00.319 --> 00:25:02.960
Thank you very much, Eileen, for bringing us that report.
00:25:03.279 --> 00:25:05.039
You're very welcome, Joe.
00:25:06.480 --> 00:25:15.519
We're having a little break again now before our next season of the Swiss Connection Science Podcast, but we'll be back very soon with a whole season with new pharma stories.
00:25:15.839 --> 00:25:24.079
In particular, we'll have a story of parents becoming unexpected drug developers to save the lives of their children suffering from a rare disease.
00:25:24.400 --> 00:25:28.160
Hit subscribe now in your podcast feed to make sure you don't miss out.
00:25:29.759 --> 00:25:40.240
For more science stories, visit our website swisinfo.ch, and you can help others to find our podcast by leaving us a five-star review or just share our podcast with a friend.
00:25:45.440 --> 00:25:50.720
Today's episode was recorded and edited by our science and video journalist Micheli Andina.
00:25:50.960 --> 00:25:51.759
I'm Jo Fay.
00:25:52.000 --> 00:25:53.200
Thanks for listening.