WEBVTT - Scanning Every Muscle to Help Olympians Get Stronger 0:00:08.240 --> 0:00:19.440 Hushkin, what's one surprising thing your work has taught you 0:00:19.600 --> 0:00:21.079 about elite athletes? 0:00:22.160 --> 0:00:24.680 I never thought I would see muscles that were so 0:00:25.560 --> 0:00:30.520 developed they broke our scale. Wow. 0:00:31.960 --> 0:00:34.159 Yeah, like it was just too big the machine, the 0:00:34.240 --> 0:00:35.839 AI couldn't figure out what it is. 0:00:35.960 --> 0:00:38.159 Well, no, the AI found it, but we are like 0:00:38.240 --> 0:00:39.600 our kind of rating system. 0:00:40.200 --> 0:00:43.280 Wow. Was there a particular athlete or a particular sport 0:00:43.360 --> 0:00:44.400 or particular muscle? 0:00:44.520 --> 0:00:44.680 What? 0:00:44.680 --> 0:00:46.320 What? What muscle broke the scale? 0:00:47.400 --> 0:00:49.720 Uh? The gluteus maximus breaks it A. 0:00:49.760 --> 0:00:53.040 Fair kidding fantastic. 0:00:53.120 --> 0:00:54.639 Yes, it's a pain in my butt. 0:00:56.800 --> 0:00:58.360 Like because it's too big. 0:00:58.880 --> 0:01:00.720 Yeah, it's just so big. 0:01:06.680 --> 0:01:09.440 I'm Jacob Goldstein and this is what's your Problem? This 0:01:09.600 --> 0:01:12.520 month a bunch of Pushkin podcasts are coming out with 0:01:12.680 --> 0:01:17.839 Olympics inspired shows. Revisionist History has a series about America's 0:01:17.840 --> 0:01:22.640 decision to participate in Hitler's Berlin Olympics in nineteen thirty six. 0:01:23.200 --> 0:01:25.960 The Happiness Lab has an interview with a coach who 0:01:26.000 --> 0:01:29.959 coaches coaches and here on What's Your Problem, We're going 0:01:30.040 --> 0:01:32.200 to be talking with people who are working at the 0:01:32.240 --> 0:01:38.560 frontiers of technology to help elite athletes perform better. For example, 0:01:39.360 --> 0:01:43.760 today my guest is Sylvia Blemker. She's a professor of 0:01:43.800 --> 0:01:47.760 biomechanical engineering at the University of Virginia, and she's the 0:01:47.800 --> 0:01:52.200 co founder of a company called Springbok Analytics. Sylvia's problem 0:01:52.320 --> 0:01:55.880 is this, how do you combine MRI scans and artificial 0:01:55.920 --> 0:01:59.560 intelligence to generate new insights that can help both elite 0:01:59.600 --> 0:02:03.440 athletes and people suffering from diseases that affect the muscles. 0:02:04.280 --> 0:02:09.400 Springbox clients include medical researchers, Olympic athletes, major League baseball, 0:02:09.760 --> 0:02:13.920 and several professional basketball and soccer teams. You'll hear about 0:02:13.960 --> 0:02:16.440 all that on the show, but first we're going to 0:02:16.520 --> 0:02:20.000 pick up where we left off in the conversation. We 0:02:20.000 --> 0:02:25.480 were discussing the extraordinarily large muscles of elite athletes, and 0:02:25.520 --> 0:02:28.840 then Sylvia told me something even more surprising. 0:02:30.000 --> 0:02:32.960 The other thing is that they have some tiny muscles too, Huh. 0:02:33.360 --> 0:02:36.839 Like they have like smaller than a normal person's muscle. 0:02:36.880 --> 0:02:37.440 Much smaller. 0:02:37.560 --> 0:02:38.040 Huh. 0:02:38.120 --> 0:02:39.639 They put their muscle where they need it. 0:02:40.120 --> 0:02:42.640 What's an example, Like what muscle is tiny and what 0:02:42.720 --> 0:02:43.440 kind of athletes? 0:02:44.320 --> 0:02:49.239 Calf muscles are small in most fast athletes huh? And 0:02:49.360 --> 0:02:53.399 ether you look at a sprinter or like a running back. 0:02:53.960 --> 0:02:57.279 It's just all quad no calf, all. 0:02:57.040 --> 0:03:00.880 Like thigh no calf, yeah, THI And it kind of 0:03:00.880 --> 0:03:03.000 makes sense because you know, if you're trying to run fast, 0:03:03.000 --> 0:03:04.840 you wouldn't want to put a lot of mass like 0:03:04.960 --> 0:03:07.240 at the end of your leg. It's like as a 0:03:07.240 --> 0:03:08.959 lot of inertia to like move your leg. 0:03:09.240 --> 0:03:09.520 Huh. 0:03:09.960 --> 0:03:12.880 Because you know, the muscles are important for sprinting, that's 0:03:12.919 --> 0:03:16.440 the interesting thing, but they just don't they're small, very. 0:03:16.600 --> 0:03:20.520 H huh uh huh. So I'm particularly interested at this 0:03:20.680 --> 0:03:27.679 moment in the sports piece of what you do. I'm curious, 0:03:27.800 --> 0:03:31.160 by the way. Do you work with any Olympic teams 0:03:31.240 --> 0:03:33.520 or Olympic athletes? Yeah? 0:03:33.639 --> 0:03:38.040 Yeah, We've actually been working with several different Olympic athletes. 0:03:39.440 --> 0:03:42.000 The ones that probably that come to mind most are 0:03:42.160 --> 0:03:45.760 multiple players on the US women's national soccer team. 0:03:46.000 --> 0:03:48.720 Oh cool, tell me, like, tell me the story of 0:03:48.760 --> 0:03:52.440 that of that work. So they came to you, what 0:03:52.480 --> 0:03:54.160 did what did they what do they want when they 0:03:54.160 --> 0:03:55.800 came to you? Like, how did that? How did that begin? 0:03:57.840 --> 0:04:01.240 They came to us along with their team. So the 0:04:01.280 --> 0:04:04.200 technology we provide, you know, an athlete could understand it, 0:04:04.240 --> 0:04:07.760 but really with their team to help them figure out 0:04:07.840 --> 0:04:10.120 how to keep athletes healthy. 0:04:10.320 --> 0:04:12.920 So what did they what did they say? What did 0:04:12.960 --> 0:04:13.960 they say when they came to. 0:04:14.880 --> 0:04:19.679 So, for example, one athlete that's coming to mind had 0:04:20.560 --> 0:04:24.960 a known imbalance side to side that based on a 0:04:25.120 --> 0:04:28.400 history of injury, and they really wanted to know where 0:04:28.400 --> 0:04:29.960 that imbalance was coming from. 0:04:30.080 --> 0:04:32.960 So the woman had had hurt one of her legs, 0:04:33.000 --> 0:04:35.480 and that leg was even after she came back, that 0:04:35.560 --> 0:04:38.320 leg was weaker essentially than the other. I mean, is 0:04:38.360 --> 0:04:40.960 that the sort of gross you know, macro. 0:04:41.040 --> 0:04:43.520 Way, Yeah, exactly, that's a that's a nice way to 0:04:43.520 --> 0:04:43.720 put it. 0:04:43.800 --> 0:04:47.240 Yeah, And and they wanted a sort of finder like, okay, 0:04:47.240 --> 0:04:49.359 but we can see that, but what's going on on 0:04:49.400 --> 0:04:53.200 the inside, like muscle by muscle tell us that, yes. 0:04:53.080 --> 0:04:55.800 Exactly, that's precisely what we do. We go on the 0:04:55.839 --> 0:05:00.760 inside because on the outside you see perhaps that her 0:05:01.120 --> 0:05:04.640 knee extents or quads seem weaker on one side than 0:05:04.680 --> 0:05:08.719 the other. But there's four quads, quadre steps, four muscles, 0:05:09.160 --> 0:05:11.400 and so it's not clear which of those muscles are 0:05:11.440 --> 0:05:15.839 actually the culprit for that imbalance and in what way. 0:05:15.560 --> 0:05:19.760 Good So this is their question, and then what happens next? 0:05:20.160 --> 0:05:23.960 So This first step is an MRI scan, and so 0:05:24.960 --> 0:05:32.000 with these athletes or teams, we have ways to connect 0:05:32.040 --> 0:05:35.480 them with an MRI machine, whether it be through an 0:05:35.480 --> 0:05:39.400 imaging center that they partner with, or we've even actually 0:05:39.400 --> 0:05:45.440 brought MRI mobile trucks to sites to make it. 0:05:45.040 --> 0:05:47.760 Like players run off the field and get an MRI 0:05:48.000 --> 0:05:51.200 and go back and keep playing. Yeah, yeah, kind yeah, yeah. 0:05:51.520 --> 0:05:54.440 It helps just with the timing of things. But so 0:05:54.520 --> 0:05:57.960 first we connect them there, so it takes about ten minutes. 0:05:58.560 --> 0:06:03.000 Then they send those pictures up into the cloud into 0:06:03.080 --> 0:06:06.560 our server and then we crunched through it and then 0:06:06.640 --> 0:06:10.680 we send back a report on their muscles. We also 0:06:10.920 --> 0:06:14.799 have what we call it interactive Viewer, and it's presented 0:06:14.880 --> 0:06:18.520 in the form of a three D model. Three dimensional model, 0:06:18.560 --> 0:06:22.120 so you actually see your own legs, the muscles and bones, 0:06:22.200 --> 0:06:26.280 your own muscles and bones that we've identified from the 0:06:26.320 --> 0:06:30.960 images going through a process called segmentation where we find 0:06:31.000 --> 0:06:33.160 all the muscles and bones and then we reconstruct them, 0:06:33.160 --> 0:06:35.040 so it's kind of like a digital twin of that 0:06:35.080 --> 0:06:38.479 person that they can see on their computer. And so 0:06:38.680 --> 0:06:41.240 along with it or a number or all these metrics 0:06:41.320 --> 0:06:47.880 that helps them understand their balance. The development or strength 0:06:47.920 --> 0:06:50.120 of the muscles and the health of the muscles. 0:06:50.520 --> 0:06:52.880 So tell me about this report, they get like, what 0:06:52.920 --> 0:06:54.360 does it? What does it say? 0:06:55.960 --> 0:06:57.920 So the basis of that is actually a lot of 0:06:57.960 --> 0:07:01.479 research that we did over many years, because you need 0:07:01.520 --> 0:07:06.400 to understand where somebody falls relative to a normal essentially 0:07:06.440 --> 0:07:09.320 to give them. Essentially, we have a scoring system for 0:07:09.560 --> 0:07:13.200 the muscles and that's based on comparing with a large 0:07:13.280 --> 0:07:18.440 data set of healthy individuals. And so we know for 0:07:18.480 --> 0:07:23.920 a given person, based on their sex, age, height and weight, 0:07:24.680 --> 0:07:27.280 how big we expect all the muscles to be. And 0:07:27.760 --> 0:07:30.320 that's through a lot of previous research. So then we 0:07:30.360 --> 0:07:34.120 can say, okay, here's where you land each particular muscle 0:07:34.520 --> 0:07:38.240 compared to this nor what we call a normative database, 0:07:38.840 --> 0:07:40.520 so we call it a spring box score. 0:07:40.880 --> 0:07:42.560 Do you do it for every muscle in the leg. 0:07:42.480 --> 0:07:46.360 Or we do it? So our primary product that we 0:07:46.400 --> 0:07:50.240 started with was every muscle in the legs essentially from 0:07:50.320 --> 0:07:54.520 belly to feet, no muscle left behind. They're all important. 0:07:54.720 --> 0:07:56.400 How many muscles are there. 0:07:56.960 --> 0:07:58.120 Thirty five per leg? 0:07:58.200 --> 0:08:01.800 So seven okay, okay, yeah, more than I would have guessed, 0:08:01.960 --> 0:08:07.840 but a lot. Yeah, all important, okay. So, and so 0:08:08.080 --> 0:08:12.000 it's basically, how strong and healthy is every one of 0:08:12.040 --> 0:08:15.800 those seventy muscles relative to baseline? 0:08:16.080 --> 0:08:19.960 Right, And then the asymmetry comes where you can compare 0:08:20.080 --> 0:08:22.280 side to side. So for each of the thirty five 0:08:22.360 --> 0:08:24.920 muscles that they exist on each leg, we can say 0:08:25.520 --> 0:08:27.960 which side is bigger, which side is smaller, and by 0:08:28.000 --> 0:08:31.280 how much? And then we also have normative values for 0:08:31.400 --> 0:08:34.080 that because we're all just slightly asymmetric. 0:08:34.320 --> 0:08:38.120 Uh huh, And presumably some muscles are more asymmetric than others, 0:08:38.160 --> 0:08:41.000 And so you want to know kind of how how 0:08:41.120 --> 0:08:46.880 asymmetric relative to baseline is this particular pair of muscles exactly? Yeah, 0:08:47.360 --> 0:08:50.520 And and so in the in the case of this 0:08:50.880 --> 0:08:53.920 uh soccer player who came to you who you know, 0:08:54.120 --> 0:08:56.560 knew knew she had some kind of problem with her 0:08:56.640 --> 0:08:58.839 quadriceps on one side, but didn't know what was going on. 0:08:59.320 --> 0:09:00.200 What did you find? 0:09:01.640 --> 0:09:05.880 We found some imbalances, and actually not just in those muscles. 0:09:06.640 --> 0:09:10.080 It turns out that, you know, it's all connected. So 0:09:11.679 --> 0:09:14.679 if you have a weakness or an imbalance and one 0:09:14.720 --> 0:09:17.000 set of muscles, usually some other set of muscles are 0:09:17.000 --> 0:09:18.200 compensating in someone. 0:09:18.360 --> 0:09:21.240 Yeah, well, it's like when you like mess up. Even 0:09:21.280 --> 0:09:24.600 if you're just a recreational athlete, right Like, if you 0:09:24.640 --> 0:09:26.800 mess up something, you mess up your ankle, then you 0:09:26.840 --> 0:09:29.280 start walking funny, and then like your back hurts because 0:09:29.280 --> 0:09:32.240 you're walking funny. Right Like, that is a very anecdotally 0:09:32.240 --> 0:09:33.000 apparent thing. 0:09:33.080 --> 0:09:35.160 Yeah, yeah, we all know that, but that you know 0:09:35.240 --> 0:09:37.440 it shows through in the skin. But the thing is 0:09:37.480 --> 0:09:41.600 that it's not very intuitive from the outside which muscles 0:09:41.720 --> 0:09:45.520 have been affected, how they've compensated, and it looks different 0:09:45.559 --> 0:09:46.680 for every single person. 0:09:46.920 --> 0:09:47.439 Huh. 0:09:47.480 --> 0:09:50.280 So that's why the report is very valuable because for 0:09:50.400 --> 0:09:54.360 that person, they know exactly which muscles are the ones 0:09:54.400 --> 0:09:56.360 that they really need to target, both the ones that 0:09:56.400 --> 0:09:59.360 they already thought maybe were an issue, but then all 0:09:59.360 --> 0:10:03.160 the other ones that showed up and they didn't really realize. 0:10:03.240 --> 0:10:05.040 And so in the case of this soccer player, was 0:10:05.080 --> 0:10:09.720 it like one particular quadrcep on one side that was 0:10:09.800 --> 0:10:11.560 like the core thing and you could figure out which 0:10:11.559 --> 0:10:12.040 one it was. 0:10:12.440 --> 0:10:14.200 There was a few mess It wasn't just that. I 0:10:14.200 --> 0:10:18.480 think there were at least one calf muscle and then 0:10:18.679 --> 0:10:22.319 some in especially in the deep hip, those were impacted. 0:10:23.240 --> 0:10:26.199 So yeah, it kind of shows up everywhere. 0:10:26.679 --> 0:10:30.000 And so you have this essentially diagnosis, right, a very 0:10:30.040 --> 0:10:33.679 sort of fine grained kind of diagnosis. Do you also 0:10:33.840 --> 0:10:35.920 have a have a prescription? Do you have sort of 0:10:36.240 --> 0:10:41.280 particular kinds of training to address these very fine grained things, 0:10:41.360 --> 0:10:44.040 or do you leave that to the trainers or whoever? 0:10:44.600 --> 0:10:47.080 We leave that to the trainers, because I think that 0:10:47.160 --> 0:10:49.800 it's also important to have all the other information about 0:10:49.840 --> 0:10:53.319 the athlete. We're not arguing that it replaces everything else. 0:10:53.640 --> 0:10:57.280 And people pair it with lots of different other types 0:10:57.320 --> 0:11:02.320 of measurements depending on the application or in the setting, 0:11:02.400 --> 0:11:06.000 Like some people pair it with let's say, metrics of 0:11:06.720 --> 0:11:11.120 jump performance. I'm shifting over to basketball here, but that's 0:11:11.160 --> 0:11:12.959 just one that came to mind, where you can look 0:11:12.960 --> 0:11:16.680 at the asymmetry about how how an athlete jumps, but 0:11:16.720 --> 0:11:18.760 then you can also compare it to the asymmetry of 0:11:18.760 --> 0:11:22.840 their muscles and get some insight. So it definitely, you know, 0:11:23.080 --> 0:11:24.640 plugs in with a lot of other things. 0:11:25.000 --> 0:11:29.760 And to what extent can trainers or you know, strength 0:11:29.840 --> 0:11:36.680 coaches develop programs that are sufficiently kind of fine grained 0:11:36.679 --> 0:11:39.840 to match the kind of fine grained findings you're having, right, Like, 0:11:39.880 --> 0:11:42.920 for example, if you find, as I understand you did 0:11:43.400 --> 0:11:48.120 that a soccer player has one particular quadricep that is weak. Like, 0:11:48.400 --> 0:11:51.120 are there workouts that target a single quadricep and not 0:11:51.160 --> 0:11:51.600 the others? 0:11:52.120 --> 0:11:52.800 Yep, there are. 0:11:53.200 --> 0:11:57.240 That's cool for whichever quadricepp you're just like, just for fun, 0:11:57.320 --> 0:11:58.160 give me an example. 0:11:59.520 --> 0:12:01.840 You know. One one way that it's very simple is 0:12:01.920 --> 0:12:07.480 using something called biofeedback. Huh. So you can measure whether 0:12:07.600 --> 0:12:11.360 you use something called EMG, which is a way to 0:12:11.360 --> 0:12:14.199 measure how much electrical activity is a muscle, and then 0:12:14.240 --> 0:12:17.280 you can see which muscles you're using for a given task. 0:12:17.840 --> 0:12:20.120 So if you give people the feedback of which of 0:12:20.120 --> 0:12:23.640 those muscles they're using and say, oh, no, you're not 0:12:23.760 --> 0:12:26.880 using this one, use this one more, that actually works 0:12:27.000 --> 0:12:27.760 very effectively. 0:12:27.920 --> 0:12:32.280 Oh really, So you can basically use your brain if 0:12:32.320 --> 0:12:36.760 you're getting the feedback to focus on which quadricep you're training. 0:12:37.000 --> 0:12:39.880 Yeah, and there's other ways you can give the feedback 0:12:39.880 --> 0:12:42.240 in other different ways, but yeah, our brains are very 0:12:42.600 --> 0:12:44.880 good at that. Once they get feedback, they're very good 0:12:44.920 --> 0:12:45.520 at learning. 0:12:46.160 --> 0:12:49.080 That's cool, especially somehow to think of what elead athletes right, 0:12:49.080 --> 0:12:52.080 because they are already presumably like super dialed in in 0:12:52.160 --> 0:12:54.920 terms of like the relationship between their brain and their 0:12:54.960 --> 0:12:57.120 body at this very elite level exactly. 0:12:57.520 --> 0:13:00.920 Yeah. The other I was going to mention a lot 0:13:00.960 --> 0:13:05.199 of players and teams use this not just one time, 0:13:05.280 --> 0:13:09.640 but overtime. So they'll get a scan, figure out a plan, 0:13:10.760 --> 0:13:14.400 work on that for maybe three months or six months, 0:13:14.440 --> 0:13:16.559 and then do another scan and see how things are 0:13:16.559 --> 0:13:20.839 progressing and adjust accordingly. So that's definitely another way to 0:13:21.000 --> 0:13:24.160 in the long term see if what they're doing is 0:13:24.240 --> 0:13:26.600 resulting in the change that they're hoping to see. 0:13:27.280 --> 0:13:29.839 So, what happened with that soccer player who had the 0:13:30.120 --> 0:13:33.960 weak quadricep and other related Yeah, No, I. 0:13:34.000 --> 0:13:39.800 Think she's doing great, like staying healthy and and you know, 0:13:39.880 --> 0:13:40.480 getting ready. 0:13:41.000 --> 0:13:42.600 Yeah. So I know you can't tell us her name, 0:13:42.640 --> 0:13:47.120 but will we see her in the Olympics this time? Great? So, 0:13:47.240 --> 0:13:48.800 as you were talking about that, I mean there was 0:13:48.840 --> 0:13:51.319 a moment where it was like, Okay, the athlete goes 0:13:51.320 --> 0:13:55.120 and gets the MRI and then you get the scan. 0:13:55.320 --> 0:13:57.400 You get the scan, and then you said, like, you 0:13:57.440 --> 0:13:59.440 crunch through the numbers and then you make the report. 0:14:00.000 --> 0:14:03.960 Ssumably you crunching through the numbers is like the result 0:14:04.000 --> 0:14:06.120 of many many years of work and kind of the 0:14:06.120 --> 0:14:08.760 core of what your company does. So I want to 0:14:08.800 --> 0:14:10.920 talk a little bit more about that and kind of 0:14:10.960 --> 0:14:15.000 how you how you got here, Like how did you 0:14:15.040 --> 0:14:16.040 come to start the company? 0:14:16.440 --> 0:14:21.840 Mm hmm? How long do I have a while? 0:14:22.000 --> 0:14:23.680 I mean it's not you know, it's not the radio, 0:14:23.720 --> 0:14:24.520 it's a podcast. 0:14:25.360 --> 0:14:27.720 You. I am a professor too, so I can go on. 0:14:27.920 --> 0:14:29.960 Let me ask you this, What was the moment when 0:14:29.960 --> 0:14:31.600 you decided to start the company? 0:14:32.880 --> 0:14:35.520 I can give you one moment and then we can 0:14:36.080 --> 0:14:36.720 do a. 0:14:36.640 --> 0:14:42.640 Couple of moments. Yeah, that's what's what a good story is, 0:14:42.680 --> 0:14:44.080 like three moments. 0:14:43.720 --> 0:14:45.680 In time, right right? Three not four? 0:14:46.440 --> 0:14:48.360 Four is a tricky number. We could do five, or 0:14:48.360 --> 0:14:52.160 we could do one. I think I think is better. 0:14:52.600 --> 0:14:55.320 Yeah. Yeah. So I'm a professor. I run a lab 0:14:56.000 --> 0:14:59.960 and form my entire career. I've been fascinated with must 0:15:00.960 --> 0:15:05.680 and how it works, and fascinated by something that we 0:15:05.800 --> 0:15:09.640 all in like the muscle field. We call form function relationships. 0:15:10.280 --> 0:15:13.080 So the idea that the way a muscle is shaped 0:15:13.600 --> 0:15:15.760 and the way it's structured and how big it is 0:15:16.360 --> 0:15:20.000 influences how well it works or how well it functions, 0:15:20.040 --> 0:15:22.760 how strong it is, how well it behaves. And there's 0:15:22.800 --> 0:15:24.640 a lot too there, and there's a lot of nuance. 0:15:24.680 --> 0:15:27.480 And that's like I've spent a career studying that in 0:15:27.520 --> 0:15:30.520 lots of different ways. So I've always been interested in 0:15:30.600 --> 0:15:33.520 quantifying muscle and figuring out how that influence is, how 0:15:33.520 --> 0:15:36.800 it works, and both in healthy people or in athletes, 0:15:36.920 --> 0:15:41.800 and also in different patient populations different In particular, I 0:15:41.880 --> 0:15:47.400 have an interest in movement disorders, so neuromuscular diseases that 0:15:47.560 --> 0:15:53.400 lead to impairments and mobility and movement ability. So one 0:15:53.400 --> 0:15:56.400 of the light bulb moments for this was the fact 0:15:56.440 --> 0:15:59.600 that I'd been using MRI to study muscle in my 0:15:59.640 --> 0:16:03.680 research for a long time. It's kind of a ubiquitous 0:16:03.720 --> 0:16:09.080 tool or like often use tool and research. But I 0:16:09.160 --> 0:16:14.160 was struck by the fact that I was hearing from, 0:16:14.200 --> 0:16:18.320 in particular a surgeon collaborator, and the surgeon was telling 0:16:18.360 --> 0:16:21.920 me about his work and helping kids with cerebral palsy 0:16:22.080 --> 0:16:26.280 improve their movement where they had hindered movement and largely 0:16:26.400 --> 0:16:31.320 because their muscles are impacted. Not only do they have 0:16:32.160 --> 0:16:35.680 an impaired ability to control their muscles, but their muscles 0:16:35.800 --> 0:16:38.840 end up with impairments in their structure or their form, 0:16:39.440 --> 0:16:42.680 which then influences how well they work. So surgeons have 0:16:42.760 --> 0:16:44.920 to go in and do surgeries to try to change that. 0:16:45.560 --> 0:16:49.200 They do things like modify tendons to try to make 0:16:49.320 --> 0:16:53.960 muscles less stiff, or they transfer muscles to make them 0:16:53.960 --> 0:16:56.960 do a new thing. But one of the big tricky 0:16:57.000 --> 0:17:00.240 parts is that oftentimes some of those muscles are very week. 0:17:00.920 --> 0:17:03.600 So if they choose the wrong muscle, then they'll make 0:17:03.600 --> 0:17:07.359 a weak muscle even weaker. Huh, And that's catastrophic. So 0:17:07.400 --> 0:17:10.280 it's a very fine line that a surgeon has to 0:17:10.280 --> 0:17:11.639 figure out and they have to go in. You know, 0:17:11.720 --> 0:17:14.399 we just talked about there's thirty five muscle muscles in 0:17:14.440 --> 0:17:16.359 each leg, So which of the muscles are the ones 0:17:16.400 --> 0:17:19.480 that should be operated on and which ones should be avoided? 0:17:20.200 --> 0:17:25.320 And so what my collaborator, a guy named doctor Mark Abel, 0:17:26.119 --> 0:17:30.440 fantastic surgeon. He was telling me, Yeah, like, it's very hard, 0:17:30.480 --> 0:17:32.480 and I don't he didn't have a way to see that. 0:17:32.720 --> 0:17:34.840 All that he could do is look from the outside. 0:17:35.280 --> 0:17:37.960 No technology could give him the information he needed to 0:17:38.080 --> 0:17:40.840 figure out which muscles he should focus on and which 0:17:40.840 --> 0:17:41.240 ones to. 0:17:41.200 --> 0:17:44.240 Avoid, because it's not obvious by looking what's a strong 0:17:44.320 --> 0:17:47.240 muscle and what's a weak one. Yeah, I guess that's 0:17:47.240 --> 0:17:52.920 surprising to me on some level, Like I don't I've 0:17:52.960 --> 0:17:55.600 never thought about it. But naively, I would think you 0:17:55.680 --> 0:17:59.080 could look at a muscle and say it looks strong 0:17:59.160 --> 0:17:59.720 or it looks weak. 0:17:59.760 --> 0:18:02.200 Not right, because you just see it from the surface. 0:18:02.320 --> 0:18:06.000 You don't see it on the inside. And the other challenges. 0:18:06.640 --> 0:18:10.879 For every joint, there's many muscles. So like we just 0:18:10.880 --> 0:18:13.439 said that quadrceps has four muscles on the back of 0:18:13.440 --> 0:18:18.040 the leg, ham strings, there's three hamstrings muscles. There's other 0:18:18.119 --> 0:18:21.480 muscles that are in the thigh. So you're just seeing 0:18:21.520 --> 0:18:24.520 what's an impairment and the overall movement, let's say of 0:18:24.560 --> 0:18:26.960 a joint. But then there could be many muscles or 0:18:27.000 --> 0:18:29.119 combinations of muscles that are leading to that, and you 0:18:29.160 --> 0:18:32.679 don't know when you look from the outside. Our body 0:18:32.720 --> 0:18:34.640 is designed that way actually to be somewhat we call 0:18:34.640 --> 0:18:37.440 it redundant. We have more muscles than we need probably, 0:18:37.560 --> 0:18:41.560 but if you think about imbalances, then any one of 0:18:41.600 --> 0:18:43.879 those muscles could create it create some trouble. 0:18:44.119 --> 0:18:48.320 So, okay, so the surgeon describes this problem he's having, 0:18:48.880 --> 0:18:50.520 and then and then what do you do? 0:18:50.840 --> 0:18:53.359 So then I was thinking, well, you know, that's the 0:18:53.400 --> 0:18:57.200 information that we generate all the time. When we're doing 0:18:57.240 --> 0:19:03.240 our research. We take MRIs, quantify, we identify muscles, we 0:19:03.359 --> 0:19:06.240 create three dimensional models of the muscles, we figure out 0:19:06.240 --> 0:19:10.719 how they're working from that. But I was struck by 0:19:10.760 --> 0:19:12.560 the fact that none of that was something that a 0:19:12.600 --> 0:19:16.320 clinician could use, despite the fact that MRI is obviously 0:19:16.440 --> 0:19:20.399 ubiquitous in healthcare. Right, yeah, you can't go to a 0:19:20.400 --> 0:19:23.639 hospital without finding multiple MRIs, but there's no way to 0:19:23.760 --> 0:19:26.000 use those MRIs and the way that I was using 0:19:26.000 --> 0:19:30.200 them for my research, And I thought, well, that's too bad, 0:19:30.280 --> 0:19:33.840 because this would be very useful to the surgeon in 0:19:33.920 --> 0:19:38.560 figuring out how to treat these patients. So that was 0:19:38.640 --> 0:19:41.439 one light bulb at the beginning. So a lot of 0:19:41.480 --> 0:19:43.439 it was figuring out how to take something that we 0:19:43.600 --> 0:19:48.240 use in research for very specific, targeted basic science questions 0:19:48.560 --> 0:19:51.560 and turn it into something that is useful clinically. 0:19:52.040 --> 0:19:55.080 And again this is like my ignorance, Like I might 0:19:55.119 --> 0:19:58.119 have thought, well, you could just do an MRI and 0:19:58.200 --> 0:20:02.280 see how big or not big the muscles are, and 0:20:02.359 --> 0:20:04.920 infer from how big or not big the muscles are, 0:20:05.800 --> 0:20:10.480 how strong or not strong they are. And that sounds straightforward, 0:20:11.920 --> 0:20:14.960 but clearly it's not, like, why is it harder than that? 0:20:16.320 --> 0:20:20.040 So a couple of reasons. One is going taking the 0:20:20.160 --> 0:20:23.119 MRI pictures and figuring out how big the muscles is 0:20:23.160 --> 0:20:27.800 a very challenging problem. So in order to accurately get 0:20:27.840 --> 0:20:31.520 how big the muscles are, you have to essentially generate 0:20:31.560 --> 0:20:33.840 its shape in three dimensions, so you have to get 0:20:33.840 --> 0:20:37.000 the whole length of the muscle, and so you do 0:20:37.119 --> 0:20:40.800 that off of multiple MRI pictures. So the MRI essentially 0:20:40.840 --> 0:20:45.520 kind of takes pictures through the body at multiple different 0:20:45.680 --> 0:20:49.960 slices we call them, going from you know, the abdomen 0:20:50.000 --> 0:20:52.600 all the way down to the feet, sort of going 0:20:52.840 --> 0:20:55.840 cross sectionally we call it. And so we usually have 0:20:56.240 --> 0:21:00.439 over two hundred images. So in each image you have 0:21:00.480 --> 0:21:03.480 to find each muscle, and so for any given image 0:21:03.520 --> 0:21:08.800 there's probably at least fifteen muscles or more. 0:21:09.359 --> 0:21:11.439 So it wasn't like you could just push the like 0:21:11.840 --> 0:21:14.440 show me the muscles button on the MRI and it 0:21:14.440 --> 0:21:15.359 would show you the muscles. 0:21:15.359 --> 0:21:15.399 Like. 0:21:15.440 --> 0:21:18.440 Nobody had done that, and there was no obvious way 0:21:18.480 --> 0:21:21.560 to do it, certainly not for a surgeon. Ordering a 0:21:21.920 --> 0:21:23.880 standard MRON just didn't exist. 0:21:24.240 --> 0:21:25.159 It did not exist. 0:21:25.560 --> 0:21:30.840 So okay, so you realize this, what happens? How do 0:21:30.840 --> 0:21:31.520 you make it happen? 0:21:32.280 --> 0:21:34.720 So, you know, one of our first tasks was to 0:21:34.760 --> 0:21:38.480 figure out how to get many muscles. So one of 0:21:38.520 --> 0:21:41.320 the things that we had done on the research side 0:21:41.400 --> 0:21:43.920 is really focus on a couple of muscles, but I 0:21:44.000 --> 0:21:46.600 knew for this application that wasn't going to work. We 0:21:46.720 --> 0:21:49.840 have to be able to identify any muscle. That was 0:21:49.840 --> 0:21:52.600 really the problem is that like you don't know which 0:21:52.600 --> 0:21:54.399 one's the problem, so you don't know which one to 0:21:54.440 --> 0:21:56.040 look at. So you got to look at all of them. 0:21:56.520 --> 0:21:59.000 And then the next task was to figure out all 0:21:59.000 --> 0:22:01.840 those muscles and figure out a process to go from 0:22:01.880 --> 0:22:04.920 the you know, identifying each and every muscle and each 0:22:04.960 --> 0:22:08.359 and every image. So it's called developing an atlas. 0:22:08.560 --> 0:22:10.840 And is that an AI problem? 0:22:12.080 --> 0:22:14.320 So now we have an AI, and it's the type 0:22:14.320 --> 0:22:18.640 of AI as supervised learning, where they can you essentially 0:22:18.800 --> 0:22:21.520 train the computer to do what the person would do. 0:22:22.000 --> 0:22:23.879