WEBVTT - Health: Stellenbosch University uses AI to help detect TB 0:00:00.160 --> 0:00:04.760 Now then, tuberculosis remains the world's deadliest infectious disease, and 0:00:05.000 --> 0:00:07.440 here in South Africa the numbers are staggering four hundred 0:00:07.480 --> 0:00:10.240 and sixty eight out of every one hundred thousand people affected. 0:00:10.960 --> 0:00:15.840 But despite this, TB remains massively underdiagnosed in our health 0:00:15.840 --> 0:00:19.160 and wellness segment. Today, we're focusing on a groundbreaking new 0:00:19.200 --> 0:00:22.400 initiative from Stellenbosch University, where researchers are helping lead a 0:00:22.440 --> 0:00:28.040 global trial using AI artificial intelligence to improve TB diagnoses, 0:00:28.200 --> 0:00:34.280 particularly at primary care level. The project, involving ten institutions 0:00:34.400 --> 0:00:38.200 across Africa and Europe, aims to develop a diagnostic tool 0:00:38.280 --> 0:00:42.760 that uses handheld ultrasound devices and smartphones to help detect 0:00:42.840 --> 0:00:47.040 TV quickly and more accurately, even in rural or low 0:00:47.080 --> 0:00:50.440 resource settings. Joining us now to explain more is Professor 0:00:50.560 --> 0:00:56.800 Grant Thron, Professor and Clinical Microbacteriology and Epidemiology at Stellenbosh Universities, 0:00:56.800 --> 0:01:00.120 also the char coordinator for this innovative study. Good to 0:01:00.160 --> 0:01:02.120 have you with us this afternoon. Thanks for making time. 0:01:02.400 --> 0:01:04.720 Thank you, mich Thank you so much. Sarah Jane, Good 0:01:04.720 --> 0:01:05.920 afternoon to you and your listeners. 0:01:06.520 --> 0:01:09.760 It's incredible to think that although it is the world's 0:01:09.880 --> 0:01:14.800 sort of most deadliest infectious disease, it's still underdiagnosed. Why 0:01:14.920 --> 0:01:15.560 is that. 0:01:15.520 --> 0:01:20.160 Rov The main reason is that TB has such a 0:01:20.280 --> 0:01:24.959 diverse way of presenting. Most TB, in fact, probably about 0:01:24.959 --> 0:01:28.560 half TB is in people who don't yet feel sick 0:01:28.680 --> 0:01:31.679 enough to seek care. These are people who do not 0:01:31.760 --> 0:01:36.399 yet have full blown symptoms, and traditionally we've unfortunately relied 0:01:36.440 --> 0:01:40.320 on symptoms to trigger TB testing. But with projects such 0:01:40.360 --> 0:01:43.160 as this one, we are trying to test people earlier 0:01:43.200 --> 0:01:45.800 and earlier before they are even actually aware that they 0:01:45.800 --> 0:01:46.880 may have TB disease. 0:01:47.360 --> 0:01:50.120 So what would and what would that look like? Then? 0:01:50.200 --> 0:01:52.240 How did how? How does that work? 0:01:52.880 --> 0:01:56.280 So how that would look like? Is that you would 0:01:56.320 --> 0:02:00.200 move away from a symptom based approach to really a 0:02:00.280 --> 0:02:04.800 risk factor based approach. Right, So, regardless of how someone 0:02:05.000 --> 0:02:08.560 is feeling, if they are in a certain area, or 0:02:08.600 --> 0:02:12.720 they have maybe a clinical condition like diabetes or HIV 0:02:12.960 --> 0:02:17.160 that increases their risk of TV, you would evaluate them 0:02:17.400 --> 0:02:21.040 using a two step process. The first process is really 0:02:21.160 --> 0:02:25.720 just working out would that person actually benefit from TV testing? 0:02:26.320 --> 0:02:28.880 And that is called a screening tool where you take 0:02:28.919 --> 0:02:32.639 someone who is healthy and you see, do they actually 0:02:32.680 --> 0:02:38.200 need that crucial yet really expensive follow on commatory testing 0:02:39.520 --> 0:02:41.920 or can they go about dead with their day and 0:02:41.960 --> 0:02:45.200 we can be confident they don't have TV. So this 0:02:45.320 --> 0:02:48.400 project is really aiming to target people in those risk 0:02:48.440 --> 0:02:52.360 groups to make sure that we get the scarce but 0:02:52.720 --> 0:02:57.440 expensive tests that we have applied to the right combination 0:02:57.560 --> 0:02:59.520 of people in an efficient manner. 0:03:00.040 --> 0:03:03.560 For a while testing the wrong people at the wrong time. 0:03:04.720 --> 0:03:07.680 I suppose that's what you were saying just then, is 0:03:07.720 --> 0:03:12.960 that this is about getting to people before the point 0:03:12.960 --> 0:03:15.919 at which they're presenting with symptoms exactly. 0:03:16.040 --> 0:03:19.400 This is one of the paradoxes in TB care is 0:03:19.400 --> 0:03:22.160 that most people who we test for TB do not 0:03:22.320 --> 0:03:25.400 actually have TV. But at the same time, we just 0:03:25.560 --> 0:03:28.880 never test and laugh people for TV. So we need 0:03:28.919 --> 0:03:32.679 to make TB testing itself more efficient, so make sure 0:03:32.760 --> 0:03:37.800 that we don't waste unnecessary resources testing people who would 0:03:37.840 --> 0:03:40.880 probably not benefit, and then at the same time use 0:03:40.960 --> 0:03:44.440 those freed up resources to expand the knit for TV 0:03:44.520 --> 0:03:48.600 testing as wide as possible. But in people who have 0:03:48.760 --> 0:03:52.000 those risk factors that mean that the odds are finding 0:03:52.040 --> 0:03:54.120 someone are the highest that they can be. 0:03:54.760 --> 0:03:58.160 You mentioned some people who are more predisposed than others. 0:03:58.200 --> 0:04:00.400 Who else might be on that list who is particularly at. 0:04:00.320 --> 0:04:06.200 Risk for TV, So it depends, right, I think the 0:04:06.240 --> 0:04:10.600 most important thing is traditionally people who have some form 0:04:10.680 --> 0:04:14.960 of compromised immunity are at risk of TB. But crucially 0:04:15.200 --> 0:04:18.960 that is not just HIV. It can also be malnutrition. 0:04:19.640 --> 0:04:22.880 It can also be poor health for other reasons because 0:04:22.880 --> 0:04:26.719 of smoking or you're exposed to pollution. It can also 0:04:26.800 --> 0:04:30.520 be the fact that you maybe had TB before and 0:04:30.880 --> 0:04:34.719 that itself increases your risk for future episodes. So the 0:04:34.839 --> 0:04:38.960 risk factors for TB are diverse. But having said that, 0:04:39.040 --> 0:04:42.920 I'm really much just emphasized is that TB, even though 0:04:42.960 --> 0:04:45.840 it's more likely to be found in certain people, there 0:04:45.839 --> 0:04:50.480 are many people who themselves lead act active, healthy, normal 0:04:50.520 --> 0:04:53.760 lives who actually start to come down with a cough 0:04:53.920 --> 0:04:57.400 and often TB is not suspected in those people because 0:04:57.520 --> 0:05:02.120 they don't fit that stereotypical picture. But they often turn 0:05:02.200 --> 0:05:05.360 out to have had full blown TV all along. So 0:05:05.920 --> 0:05:10.680 there's nothing absolute with TV. It can affect everyone to 0:05:10.880 --> 0:05:11.719 different extents. 0:05:11.839 --> 0:05:16.239 Yeah, South Africa is now part of this major international project. 0:05:16.320 --> 0:05:18.680 So what's involved in the study and what exactly is 0:05:18.760 --> 0:05:19.599 the main goal. 0:05:21.400 --> 0:05:24.839 The main goal of the study is to catalyze on 0:05:25.080 --> 0:05:30.919 recent advances in ultimo ultrasound technology. So ultrasound technology which 0:05:32.120 --> 0:05:34.120 many of us are may be familiar with when it 0:05:34.160 --> 0:05:39.680 comes to you know, ultrasounds in trolleys on hospitals that 0:05:39.720 --> 0:05:44.240 are maybe used for looking at pregnancy. Those devices have 0:05:44.360 --> 0:05:49.600 become miniaturized and have become handheld. They can be moved 0:05:49.800 --> 0:05:55.400 over the body to acquire images to look for small 0:05:55.960 --> 0:05:59.640 early signs in the anatomy of people for TV. And 0:05:59.680 --> 0:06:03.719 where AI comes in is that traditionally to acquire those 0:06:03.800 --> 0:06:08.760 ultrasilved images is a lot of information that's generated and 0:06:08.960 --> 0:06:12.520 there is a very high level of training and specialized 0:06:12.560 --> 0:06:17.640 expertise that is required to analyze that large amount of information, 0:06:17.760 --> 0:06:20.520 and that is quite frankly expertise, but we do not 0:06:20.680 --> 0:06:23.120 have available in South Africa at the scale that we 0:06:23.200 --> 0:06:26.599 need it. So the AI comes in and being able 0:06:26.640 --> 0:06:30.800 to process that information as it is captured by these 0:06:30.839 --> 0:06:34.360 point of care ultrasolt machines to say that this person 0:06:34.440 --> 0:06:38.520 perhaps has a shadow on their legs, or there's evidence 0:06:38.600 --> 0:06:44.240 of tissue damage in their body, and whether or not 0:06:44.440 --> 0:06:48.479 that specific signal or that specific type of damage that 0:06:48.600 --> 0:06:54.320 is seen is itself indicative of TV And so this 0:06:54.400 --> 0:06:58.359 project will try and show that as a proof of 0:06:58.440 --> 0:07:01.920 concept and generate the evidence to say that this is 0:07:01.960 --> 0:07:04.960 a technology or maybe it would not be a technology 0:07:05.279 --> 0:07:09.559 that really requires large scale adoption because it helps us 0:07:09.680 --> 0:07:14.000 to tist more people and importantly, it may not be 0:07:14.080 --> 0:07:19.360 constrained by the huge shortage of people with irrelevant training 0:07:19.920 --> 0:07:22.000 mast the health workers in South Africa that are in 0:07:22.040 --> 0:07:23.280 such a shortage. 0:07:23.560 --> 0:07:26.920 So I was wondering how this could then, what would 0:07:26.960 --> 0:07:31.440 this sort of look like in local clinics, And I 0:07:31.480 --> 0:07:34.560 mean how soon this type of technology might be available 0:07:35.080 --> 0:07:38.160 to be rolled out in local clinics, But how would 0:07:38.160 --> 0:07:41.400 it help with with the challenges that we face here 0:07:41.520 --> 0:07:42.320 in South Africa. 0:07:43.360 --> 0:07:48.320 So the visionire is that in approximately five years or so, 0:07:49.040 --> 0:07:53.480 anyone who's coming into a clinic primary care clinic like 0:07:53.520 --> 0:07:56.720 in other words, their local community clinic, anyone who is 0:07:56.760 --> 0:08:01.720 coming in there for any reason, the seas this ultrasound 0:08:02.040 --> 0:08:05.480 scan which can potentially be done through the clothes, and 0:08:05.520 --> 0:08:07.800 it can be done by someone with only a few 0:08:07.880 --> 0:08:11.160 days of training. So perhaps while you're in the queue, 0:08:11.160 --> 0:08:14.800 while you're waiting for other clinical services, you could receive 0:08:14.880 --> 0:08:18.640 this scan of your body and that could tell you 0:08:18.760 --> 0:08:21.720 whether or not you need to be prioritized for other 0:08:21.840 --> 0:08:25.800 further testing. But that's the initial use case that we 0:08:25.880 --> 0:08:31.080 are targeting, where we are able to check anyone who's 0:08:31.080 --> 0:08:34.320 coming into the facility. And then further down the line, 0:08:34.640 --> 0:08:37.040 we know that there's a lot of TV in communities 0:08:37.080 --> 0:08:40.280 that never comes into facilities and you can maybe deploy 0:08:40.320 --> 0:08:46.360 these devices in communities where people check the torso for 0:08:46.559 --> 0:08:49.439 signs or TV and use that to refer people on it. 0:08:50.360 --> 0:08:53.800 It's really exciting stuff, isn't it. And I wonder, I mean, 0:08:53.880 --> 0:08:56.679 what if we were to look forward, what might be 0:08:56.720 --> 0:09:00.079 the long term hope for how guess AI. That's the 0:09:00.080 --> 0:09:03.720 the buzzword at the moment. How AI could transform TB 0:09:03.880 --> 0:09:07.800 care or even more broadly broader healthcare diagnostics, particularly in 0:09:07.880 --> 0:09:10.520 resource limited environments like ones that we see here in 0:09:10.559 --> 0:09:11.160 South Africa. 0:09:11.960 --> 0:09:14.600 So this is a tool that is by no means 0:09:14.640 --> 0:09:18.440 going to replace doctors or right yeah, it's a tool 0:09:18.520 --> 0:09:23.160 to simply support them where people with very specialized expertise 0:09:23.240 --> 0:09:30.960 of skits. It also is something that almost democratizers access 0:09:31.320 --> 0:09:36.280 to new diagnostic technologies because that patient to benefit from 0:09:36.280 --> 0:09:39.760 for example, the advances in ultrasound does not need to 0:09:39.800 --> 0:09:43.680 go to a university hospital perhaps where a professor who 0:09:43.760 --> 0:09:48.719 is a specialist in ultrasound is based, where easily orchese 0:09:48.880 --> 0:09:53.280 level of expertise is required to interpret that ultrasound. So 0:09:53.760 --> 0:09:59.800 it's allowed their access to new technologies by removing these 0:10:00.400 --> 0:10:06.000 bottlenix that we have in terms of human capacity and importantly, AI, 0:10:06.200 --> 0:10:10.440 even though itself is relatively complex, a lot of that 0:10:10.559 --> 0:10:14.240 complexity is behind the scenes. In other words, there I 0:10:14.360 --> 0:10:18.520 itself can be in a very user friendly tool which 0:10:18.559 --> 0:10:22.400 can simplify what would otherwise be fairly complex decision making 0:10:22.880 --> 0:10:27.880 and also importantly help frontline health workers at faster with 0:10:28.480 --> 0:10:29.760 also greater accuracy. 0:10:30.880 --> 0:10:33.880 RAP appreciate your time this afternoon, what a fascinating area 0:10:33.960 --> 0:10:38.120 of work and certainly looking forward to seeing how that 0:10:38.480 --> 0:10:42.559 is rolled out across South Africa. Speaking to Professor Granthron, 0:10:42.679 --> 0:10:47.800 professor in Clinical Microbacteriology and Epidemiology at Stelleinbosh University talking 0:10:47.800 --> 0:10:51.920 about this project that researchers at Stelleinbosh University are involved 0:10:52.440 --> 0:10:58.000 in leading really a global trial using artificial intelligence to 0:10:58.120 --> 0:11:02.520 improve tuberculosis diagnosis. As we heard there, the idea that 0:11:02.520 --> 0:11:06.040 that could be rolled out at primary care level. Number 0:11:06.080 --> 0:11:09.720 of institutions across the continent and in Europe aiming to 0:11:09.800 --> 0:11:14.800 develop a tool that essentially is a handheld ultrasound device 0:11:15.600 --> 0:11:20.400 and which could help detect TB more quickly and more accurately, 0:11:20.440 --> 0:11:24.000 particularly in rural or low resource settings. What a fascinating conversation.