[DAVID SIMON]: Okay, hello, everyone, and welcome. Thank you for attending Challenges for Mobile Diagnostics: Mobile MRI as a Case Study. My name is David Simon; I'm a research fellow at The Petrie-Flom Center for Health Law Policy, Biotechnology, and Bioethics at Harvard Law School. The Petrie-Flom Center is delighted to bring you this event as part of our
research project diagnosing in the home the ethical legal and regulatory challenges and opportunities of digital home health. Before we get to the discussion a few housekeeping matters, we welcome audience questions for our speakers, so please submit your questions. Throughout this entire event we will be pulling those questions to pose to the panelists. You may be asking: how do I submit questions? The best way to do it is to use the Zoom Q&A feature, which if you scroll towards the bottom of your zoom window you'll see a little button marked Q&A; just type your question in there. I'll be checking it very frequently. You're also welcome to join the conversation or submit a question on Twitter using the #privatemris; again that's hashtag #private M as in mike, R is in roger, I is in iris, s as in sam. If you do submit a question there, Petrie-Fom staff will be monitoring it and we'll pull it into the zoom Q&A feature. Ways that you should not try to submit questions: the raise your hand feature on Zoom;
we will not be checking that feature. We've also turned off the chat function so that our panelists can really focus on what's going on, so that's not going to be a venue for you to submit questions. If you're interested in this event and interested in other health policy, bioethics, biotechnology related topics, we strongly encourage you to sign up for the Petrie-Flom center newsletter to read the blog the Bill of Health which features some really cutting edge commentary by legal scholars.
Before I introduce our panelists, I want to thank some of our Petrie-Flom staff who helped make this event possible: Laura Chong and Chloe Reichel. Thank you very much. Our panelists today are Dr. Damien Fair, professor at the Institute of Child Development, department of pediatrics at the University of Minnesota medical school, and he's also the Redleaf Endowed Director for the Masonic Institute for the Developing Brain. We also have Dr. Francis Shen, professor of law, Mcknight presidential fellow at the University of Minnesota Law School. I'd just like to say a few words to introduce the topic before we get going. Technological progress is moving at a rapid clip, of course the COVID-19 pandemic illustrates this with various entities developing vaccines at breakneck speed, but the rapid development of medical technologies is not limited to vaccines. Companies have been working to develop various diagnostic tools that can be deployed in new settings using new technologies with new capabilities that were unimaginable or at least highly speculative only a decade or two ago.
For example the butterfly IQ plus is a highly mobile ultrasound device that can be used with simply an iOS or android device like a smartphone or a tablet. Another example is the EKO a digital stethoscope that enables physicians to make cardiac assessments that are recorded, playable, and analyzable on an iOS or android device, and it can also be used as a single-lead electrocardiogram. Both technologies make certain diagnostic tools, like ultrasounds and ECGs, more portable and accessible potentially allowing diagnostic assessments outside the traditional confines of the hospital, clinic or physician's office. Technological advances, however, also have limits, and moving technology outside of the clinic poses legal, ethical and social challenges for clinicians and researchers alike; for example, if testing new technology requires introducing and explaining the technology in a foreign language how can we ensure that research participants understand both the technology and what they are agreeing to? What ethical obligations do researchers have to communicate incidental findings, that is, findings they might that might occur through their research to their research subjects? Do they have an ethical obligation to provide care to their research subjects or to connect them with care? Today we have experts who will help us understand these issues and many more using a different technology, one that historically has been much less portable than the two that I mentioned, magnetic resonance imaging or MRI. With that, I'd like to turn it over to our panelists starting first with Dr. Fair.
[DAMIEN FAIR]: Alright, thank you, and thank you for having me. I'm happy to come and give you a little perspective from my space on brain imaging in the era of big data. I am what they call a cognitive neuroscientist, and I do a lot of imaging trying to understand the basics of brain development with non-invasive techniques like MRI, and I think that we've learned you know quite a bit over the last decade or so. which as I was just noting earlier. is kind of changing the landscape about how we utilize MRI for both our research and for our various types of clinical applications. Okay so there's two main bullets here for what I'll describe today, one is I'm going to give you a brief history of cognitive neuroscience and functional MRI, which i'm going to utilize as an exemplar here for where the space is going with regard to using techniques like MRI and the reason why we can see the use of these technologies in the home and other places making much more access accessible in the in the near future, and then I'll talk about some the idea of big data which is a big topic not right now, where we have many many subjects and studies and what that's taught us about where we're going. All right so the the actual term cognitive neurosciences is now about 50 years old, and it was meant to describe the intersection of the biology of the brain with various types of concepts of the mind and psychology.
There was a huge boost in this space, in this field, with the advent of PET imaging, and also what I'll talk about here, functional MRI which essentially is measuring the intricate nature of how neurons in the brain kind of relate to metabolism and blood flow which allows us to see activity in the brain related to neurons without actually ever even touching it. Now when this technology first came out and this is kind of how I started my career there was the potential of these technologies and how it could transform our mental health and clinical practice and research was was enormous; in fact I found this old paper around the time when I was just starting graduate school which was talking about how fMRI can be used — here's just a little quote: "Discussions of the future of fMRI have conjured up visions of mind reading devices used everywhere from the front door at the airport terminal to back room corporate personnel offices. At least one neuromarketing research firm is already trying to use fMRI to probe what consumers really think about their clients' product." So the potential at least for for early on was very high,
but while the techniques continue to be valuable characterizing activation patterns, its clinical utility has probably been relegated to pre-surgical planning and and even then it's not really widely used. Well in 1995 this guy by the name Brock Biswald looked at different types of activity in the brain; instead of activity related to when you're actually doing something — so you know when I'm like pressing my finger or not and looking at where the activity in the brain lies — he was looking at spontaneous activity when the brain's at rest, not doing anything at all, and what he's able to find is the spontaneous brain activity um when you're not actually doing any tasks can largely mimic the network in the structure of the brain when you're actually doing a specific task. It's very important because it allowed for the use of these techniques for people to be sitting in the scanner and looking at the function without them doing anything at all even being anesthetized, sleeping, anything of that nature, so there's many fundamental properties of organizations that have come out of these techniques over the years and the thoughts around the clinical and applied revolution of what this is called functional connectivity MRI was probably even greater, maybe rivaled the traditional MRI at the time. Still we haven't really gotten there. Well along the way, as we kind of expanded our understanding of brain organization and development using these types of techniques and research, the data sets were also increasing in sample size or the amount of data collected per subject, so now we're starting to do studies instead of traditional studies of 50 or 100 people now they're rivaling thousands of people, so 1200 people, 1500 people, 10,000 people a new study out of the UK, called UK Biobank, 100 000 people with the MRI scans, but at the same time as this started to grow, we started seeing signs of reproducibility failures where you might see something in one study in one MRI which you don't see in another study from a different institution.
Lots of papers highlighting some big data challenges and some findings that suggest that maybe how good we thought we were doing at identifying some of these characters in the brain were not actually real and not really replicable. So just to conclude this first part here, the field continues to evolve as data collected on the broader populations at a very fast rate; however, the arrival of these and these very large data sets are potentially highlighting finally some of the parameters and the context — which was required to get more reliable kind of data from from various types of studies and in the clinical ending to use more broadly in the clinical sciences more effectively, and in the second I'm just going to give you an example of what I mean and just talk about this idea of big data many subjects, so this actually comes this content actually comes from this paper that we just recently published it's in press in Nature right now; it's a really big deal for how we think — how we utilize, contextualized data from MRI, and it's called Toward reproducible brain wide association studies, and it's just a very basic question, right, and the question was: does the reliance on on typical neural energy sample sizes in our research studies provide an explanation of why, when we do large studies relating to clinical outcomes and things like that, don't really replicate and if so why is that? I'm not going to go through all the data, but I'm just going to show you a quick little example which revolves around this idea of sampling variability which, from all your very basic statistical classes, you've definitely gone over this which just measures the effect size estimate and how it varies between different sample samples from the population, and one of my colleagues says you know it's objectively boring boring and rarely considered, but it's really extremely important how we think about our findings, so here's just the example: my question was what's the relationship between height and age as as kids grow from from nine to 10 to 11 years old? Now I can go into a population — this is actually real data — and I can grab a sample of 25 people, 25 kids, and I'll see there's a relationship between the height and age — maybe it's a correlation around 0.85 or something like that, but I could go back into another population, or just try it again and grab another 25 kids, and I might get a sample that is highly variable and where the correlation across height and age is actually zero which gives me a completely different result. I can do this subsampling and sample 25, sampling 25, sampling 25, over and over and over again, and I'll get a distribution of that relationship, and what you'll see, again this is actually real data from what's called the ABCD study of adolescents, is that you see of that small sample there's lots of variation of those samples that you can get out this is sampling variability we can get correlations of height versus age order approach one or you might even get something that actually negative relationship is a completely wrong answer, and then I can redo that thing, I can redo that same exercise that samples of 40, 50, 100, 500, 9000 and look at those distributions, and what you'll see is that is that you need lots and lots of participants to be able to get the true relationship which is approximately 0.5 or so
to be able to get where the variability of what you get is much lower. And why is this important? Well it's because if I don't do that; I don't have a large enough sample, and I sample just a small, small group of people I can get findings from one research institution or one clinical research problem that shows me one answer — this is real data of brain imaging — that you might have a positive relationship between some type of some cognitive ability or psychopathology or somebody else another institution will get something that could be opposite which is even negative it means the sample size need to be much larger than we ever had imagined, and they suggest that these that that these consortium level data for many types of questions that we are trying to utilize for clinical applications need to have thousands and thousands of thousands of people to be able to get there, nearly 2000 people for some of the highest effects. Now what this kind of reminds us of actually is a place that genetics was between 10 and 20 years ago where they were identifying that lots of findings in genetics you know and how it relates to specific types of mental health disorders and other types of diseases were not really replicating either, and they also found the same type of issue, that in order to get replicable findings you need to have depending on the size of the effect you need to have thousands of participants yet all the studies were had had much fewer participants in them, which is probably related to some of the reliability issues that we have, and this is a quote actually from this one of these original papers just 10 years ago which highlights how in this new era of big data and small effects a recalibration of views about what groundbreaking findings is actually important and necessary. So of course the genetic world didn't sit there on their hands they came up with these these ideas to be able to to kind of leverage very large sample sizes to identify the relationships with the specific genes in various types of complex behaviors. Now we're not going to go through all these slides, in part because of timing, but I'll just point out that what was required is you would do what's called these polygenic risk scores where you take a bunch of findings in the genes, relate them to your outcome, and do that over thousands tens of thousands, 20 thousands, hundreds of thousands of people and then combine all those small effects to identify in the target population how they might be at risk for certain problem or a certain issue or certain mental health disorder things like that. Well in in the neuroimaging world the exact same thing is now beginning to be applied
where you can take those types of small effects or differences, apply them to these very very large samples, and then once you have all those types of risks across the brain instead of your genes, you can identify really specific relationships or risk factors related to complex behaviors like, in this case um in this case, ADHD, but again the point is that to do this correctly you need many subjects, thousands and thousands and thousands of people, to do this. Now one of the other lessons learned from this time was that is that if you don't do this right and you know the special sauce of kind of what needs to be required to be able to utilize some of this information in the most optimal way for certain types of issues that, when you're generating those big sample sizes that serve as the base of the research, is that you need to be maximally inclusive, so here's just an example of some of these polygenic risk scores in genetics and how most of those base samples were based on folks from European descent and the predictive accuracy, how well they work in folks who are not part of that initial base, seems to works extremely well if you also have the same same cultural and genetic background but not so well if you're from any different kind of cultural group. Highlighting that, this is an important lesson that we that to learn is that if we're if we're going to utilize some of these new techniques and this new understanding of how to maximize the efficacy and the applicability of rMRI and other MRI techniques then as we start generating these large data sets you need to be maximally inclusive of folks from various types of backgrounds. So the potential for non-invasive MRI to improve our understanding of brain function and clinical outcomes of brain based disorders is really, at this stage, higher than it's ever been before, but it's going to require larger samples than previously managed to really realize that potential. Technologies that make MRI or similar non-invasive neural imaging more accessible and a broad use will undoubtedly be part of this future; that's a bit of what Francis is going to talk about and why we're here, because now MRI is being made accessible, even in the home, using various types of mobile technologies, but while the work can put us on a more solid footing with regard to the fundamental findings in brain organization in clinical applications, its growth and accessibility outside of our universities and hospitals, which has been the primary target of where we collect these types of data will require a keen eye to maximize representation of the types of data collected but also an infrastructure to promote its ethical use. So I'll in there — not
exactly sure how we want to take questions but just thank you for having me. I'm glad we get to have this discussion here's just a bunch of the people involved with some of the data I just showed in the lab and also just, of course, there's lots of funding that we that we get to conduct a lot of the work. I'll stop there and and hand it over. [SIMON]: Thanks so much, Dr. Shen.
[FRANCIS SHEN[: Damon, if you stop your screen sharing, I'll share mine and we'll get going here. Well, as Damien said it's really nice to be here, thanks to Petrie-Flom, to David, to Chloe, Laura, everyone for having this program. I'm gonna talk about what Damien mentioned at the very end, the advent of more portable brain imaging. I have provocatively titled my presentation brain scans for everyone, but I want to talk about the ethical, legal, and social implications and the equity, diversity and inclusion challenges that accompany this move towards more inclusive and more pervasive brain scanning, and I'll take you on a little tour of work that we're doing including work that a working group is doing involving Damien as well. Just no disclosures; I want to acknowledge funding
from NIH and others and particularly want to acknowledge my colleagues Frances Lawrenz, Susan Wolf, Mike Garwood and the NIH for the grant work that is fueling most the presentation today; we have a grant on highly portable and cloud-enabled neuroimaging research, confronting ethics challenges in field research with new populations, and we have an awesome working group, all these folks here, so anything I say today I've learned from them, but what I say they are my own thoughts and should not be subscribed to the entire group, and if you get interested in this work, we've got a really cool website neuroimagingethics.org including his bibliography, and we put all of our work on there for everyone for free. All right, so I want to cover three things today, the first is kind of to set the stage and talk about the developments in more mobile and portable MRA, sort of but the potential even for an at-home almost concierge service MRI, and then I really want to focus the bulk of my conversation around this identifying ELSI and EDI challenges; there are many, and I'll talk about some of them and then finally just a couple words at the end towards solutions and invite you to be a part of that that conversation as well. All right, well let's start with the fun stuff which is the emergence of these new technologies.
I think it can be summarized in three, again provocative, but I think grain of truth headlines, one is brain scans for everyone including vulnerable populations. The article that Damien references, a great article and from the abstract, I was just looking over again, here's what they say: brain behavioral phenotype association stabilize and become more reproducible with sample sizes of n greater or equal to 2000. So for anyone who knows a brain imager, here's a fun game find that brain imager friend and ask what's the largest sample size of subjects in your study your MRI study; they're not going to be anywhere even in the ballpark of 2000s, which tells you for research purposes you need more brain scans and of course for clinical purposes access, so brain scans for everyone. Well in order to get that you aren't going to have radiologists everywhere; you
need to reduce the barriers of entry, and the idea is that anyone can scan, homage to a great movie a Ratatouille anyone can cook, but here anyone can scan potentially, and I'll talk about this, those who may not have the requisite training, that is, just because you can get behind the car and press the pedal and drive doesn't necessarily mean you should be driving, and finally in order to produce more brain scans, and if anyone can do it, boy, you can really take the brain scanner places it's never been before, so brain scans everywhere. These technologies on this screen they come from an article that we published last year in Neuroimage, with permission, and I'm not going to talk about any one particular technology. I'm not going to get into the details; I just want to highlight that there are a suite of new technologies — these are only some of them — different ones being developed; we'll talk a little bit, and you see some images of this is called the Hyperfine a company has a device. Mike Garwood and colleagues Tommy Vaughn are developing a different looking device; Larry Wald and MGH have a device. There are others pictured here as well, so there are many different types of devices. From a law and regulatory point of view, rather than key on any one particular piece of technology because we don't know who will become market dominant or what new technologies will show up, the thought is to find the key features, the common features, that define this suite and to prepare and anticipate them. So let me just
give you some from the headlines, so this is not sci-fi; this is Twitter and YouTube. One of the most interesting things for this conversation today is the work of professor Shaun DeonI at Brown and the Gates Foundation, and this is from his twitter feed, advanced baby imaging. They've live streamed, and put on YouTube, the first home-based MR; they have an MR in a van, and these are screenshots from their video which you can go on and watch, and this is the picture from like walking out of the house, and here's the research team right there on your front lawn, and this is from what they want to do — this is the first time they ever did it — and they want to do more they want to do MRI house calls, right. Imagine DoorDash showing up except instead of Chipotle coming out of the van, you can go in and get an MR scan, so this is really happening. It's also happening worldwide; this is just one example from the head of the Hyperfine group, and just to give you a sense of how different this is, this was a tweet on the receiving end; they said, we needed to construct a whole building for our first MRI; this one fits in a cupboard.
This is really different. Hospitals in the US are doing this, too; portable MR opens up a world of possibility; this is for our colleagues actually here at University of Minnesota M Health Fairview; this was one in Canada: new portable MR has the potential to change the future of health care. And just from two weeks ago, there's a story of a group in Tennessee that is exploring the use of this device in an ambulance. This stuff is really happening, and there are a lot of possibilities, new ones, some that we're exploring. Damien and I put in a grant — the initial one wasn't funded, but I think we'll find funding — we wanted to create a Minnesota mobile MRI lab. Now this
is, if you notice similarities, it's because for purposes of the grant this doesn't actually exist but we sort of labeled and said you know, what if we took Professor Deoni's, and sort of modified it and did here, and our thought was to address inequities and access both to research and then eventually care for MR, and we wanted to to do that. The big point is that to date magnetic resonance research and a lot of other types of neuroimaging research are geographically constrained, even our vocabulary use, you have to go to the research facility. Tomorrow's research and clinical practice is field based and potentially home based. Tou really could go anywhere and as David said at the outset MR can be understood as one of multiple technologies that are moving outside the hospital and into our everyday lives; there are a lot of reasons to be excited about these technologies. For instance for
consumers brain scans are in demand and at great convenience, can increase access — Damien talked a little bit about that — you can monitor participants in more real-life environments and potentially for some technology to be combined and have real-life interventions, maybe more objective data, and you can scan with much greater frequency. Typically we say have you ever had an MR not do you get your monthly MR right and so these things could all change and they're exciting; they're exciting both for clinical and research purposes, but there are a lot of ethical, legal, social implications and a lot of equity and diversity and inclusion challenges, and this has been the work of our grant and the work that I do um with with colleagues, and I want to talk about that. We had a first grant a couple years ago with Mike Garwood, Gil Gonzalez, MGH with Susan Wolf, and some of the core issues that we identified are the following: Informed consent, again you're out there, you're you're not in the hospital setting, privacy issues , I'll talk about those in a little bit, this is really big, establishing capacity to interpret and communicate data to remote participants. You can bring the scanner to the home; you can't bring, and so somewhat misleading, you're typically not going to have Professor Deoni and his entire crew there; you're going to have maybe just the tech. There's going to be extensive reliance on machine learning and artificial intelligence; we don't have maybe time to talk about that fully today, but it's important, and I want to flag it because of that there's potential bias and interpretive algorithms especially in diverse populations. If you can take
this scanner and you begin scanning in populations you've never stand in before that's great, but because you've never scanned with those populations before, because you haven't had diverse and large sample pools, because we're only now beginning to enter the era of big data neuroscience, that neuroimaging that Damien talked about, what do you do in the interim? Can we trust the data and the algorithm we currently have? Return of results is a major issue because what if you're out there, you're remote, and you find this brain scan, and there's something problematic, some structural abnormality? Again you've brought the brain scanner out far away, but you haven't brought the hospital, you haven't brought your entourage of expertise, how do you handle that? And of course access to data. I'll just briefly say that, if this was the traditional model, with everything pretty self-contained within a research facility or a hospital system, the new model is one in which the research facility is left behind. The scanner is out scanning; the data is being sent via the cloud; algorithms are analyzing it; radiologists might look at it, but that radiologist is not local and raises a lot of questions, when you've got a more geographically dispersed and culturally diverse set of participants, you have less immediate access to medical facilities, so if there's a major problem in an MR scan in the facilities at Harvard or MGH, University of Minnesota, the hospital is right over there; there's a pathway; you can just get them over, but if you're hours away what do you do? I mentioned the greater reliance on AI. There are also movements to utilize these technologies internationally, in remote and resource limited international settings. In the interest of time, I won't go into detail;
I just want to flag that, with another grant from NIH and again a number of colleagues, we put a piece out in her image last year, and I just want to highlight a couple of the take-home points that would be relevant for direct-to-consumer and home use as well: one is that there's a real concern around the therapeutic misconception, which if you're not familiar with that term, means for someone doing just research or just wellness, there is a misconception potential that it has clinical value or that the data derived from it can provides a brain health assessment, but that's often not the case, so there's a disconnect between what the consumer or the research participant expects or thinks they're getting and what they're actually getting; this could be especially problematic with very powerful brain data. We've got to ensure safety; I'll talk about that in a bit. The privacy issues are pronounced as well; there are structural abnormalities that could be of great interest to insurance companies for instance and how do we handle the flow of data, who gets it, who gets access to it, and its interpretation? I flagged AI, and of course, this incidental finance matters as well. Two of our take-home points here with this idea that, in developing guidance, ought to be looking for local partnerships and sustain local engagement and creating sustainable value now, this thing about local communities. When you transport those ideas to direct to consumer setting, it means you got to think a lot about those consumers, especially those in vulnerable populations, and here are some concerns, so kind of the flip side of the promise. So one of the promises was this is awesome for consumers, suddenly it's like click a button on your app and order a brain scan for later in the week, what could be what could be better? Well here's some perils: who's actually showing up to administer this brain scan? One of the values and the great promise of this technology is you don't have to go to years and years of training. If you know how to operate an iPad and you know how to potentially position someone in this device maybe you could acquire brain data; on one hand, that's great; on the other hand, that raises concerned about the standards required for those operating the equipment. Then there's
communication; it's one thing to get the data. Damien, I'm sure, will tell you that that data unless you have an expert to analyze it and then an expert to interpret isn't going to be — I wouldn't know what to do with it, a consumer won't know what to do with it; there's heavy reliance on this machinery of interpretation, and the language of interpretation, and because we've never had to do that outside the hospital or limited research setting, we don't have like a language to use. Genetics is a good example where there's an entire field of genetic counseling; you don't understand what those 23andme results mean; there's like an opportunity to understand; we don't have that parallel setup, something that I'm arguing we ought to have but we don't have it yet. And then is there a plan for handling those incidental findings? Oh great for me to have a brain scan — not so great because you found some tumor that I didn't know existed. As it
problematic? Can I live with it? Do I have the money to do further follow-up? Am I now living the next three weeks or three years in fear? Suddenly that on-demand convenient brain scan doesn't seem so convenient. Just one other set of concerns: it could increase access in many important ways, especially to remote and marginalized populations, but it could not, so how will the technology actually be used. Will it fulfill its potential promise will those go to issues I know NIH cares about for instance? Who's the workforce using this? Who are the intended consumers? Is this being marketed for a fancy brain club like only the high end could get this additional technology? How do we recruit and retain diverse populations? I can tell you because I've looked at this that at present there's no neural imager training for field-based research because it's never been done before, like you've never had to take your machine or never had the opportunity to take your machine into the field, and these are things that we're trying to think about and again access. So putting all this stuff together, let me think about solutions; just for a couple minutes.
One thing to say at the outset, and I should probably said at the beginning is that none of the technologies that I've mentioned, that I'm aware of, is intended as a replacement for fixed traditional MR. There are a lot of reasons for that; there are a number of things that these technologies which rely on lower fields and produce images of different quality, maybe sufficiently high quality, but different quality, there's some things you'll just never do, but there are also some things that they'll be able to do that fixed scanners have never done and that takes us back to where it began. There are going to be new markets for MR, again this these technologies at least in my view are less a replacement and more a supplement and a complement and an expansion of imaging. It really is imaging for potentially everyone, potentially anyone can scan, but should they? And brain scans could be done everywhere but will they and should they? And to me the biggest question right now is in this space there are — because it's so new um there are no real standards, and my question is will we collectively, the relevant groups — these are professional organizations, researchers, clinicians, developers, regulators, patients, research participants, consumers — develop high and meaningful standards to guide this world of brain imaging. There is a future in which we don't: there's a future in which brain
scans run wild; you intersect that with business and profit motives and you have a very problematic world. There's also a problematic world in which inclusivity and equity are thrown to the side and, in particular, that the development of these standards are not developed along with a diverse set of stakeholders. The work that we're doing in the grant, the work that we're doing in other grants, is trying to address this, and let me just say a couple words about that in closing. MR is one of a suite of new technologies that are going — consumer and digital technologies with Benjamin Silverman and others at McLean hospital; we've got a program at coming at Radcliffe institute called intimate data: ensuring equity as psychiatry embraces boundless data and AI, and these are the sorts of things that you know we're thinking about, I think ought to be thought about, issues of racial, gender, socioeconomic equity and justice and bias and in the grant that I flag at the beginning, this is our charge. We're right in the middle we're kind of starting to develop our first set of consensus guidance in the second of four years and our goal which we'll accomplish is to generate evidence-based consensus recommendations for the ethical conduct of research using these new technologies, using them in new and diverse field settings so I'm confident we'll contribute to these standards and I think that the sort of pessimistic features I mentioned won't come to bear, and I think law has a role to play so since this is in a law school, our event, let me close final slide about the rule of law.
The clinical path I think is straightforward. I mean it's bumpy and be convoluted but straightforward, right. The idea is let's get better measures of brain behavior and brain function . Let's do that in the real world; that gives us more individualized it's intimate data,
and it's delivered to your door. Why do we want to do that? Improve care, improve brain health, development of novel cures for brain diseases and disorders, but this just doesn't happen without the law; on one hand at this early stage where there's so much research needed, we need to regulate the research and we need to promote the research and promote it in ways that adhere to our values and as it moves from the research to the applied we're now boy actually we do have co-shared service; we do have this proliferation of brain data; we need new policies new standards guidelines training implementation and new laws and the work that Petrie-Flom Center does at these intersections you know is right at the heart of it. I'll stop there and stop sharing my screen and thanks to everyone for questions that I'm sure will come great thank you so much. [SIMON]: I think I'll start off with a question that picks up on something you mentioned towards the end of your talk which is the market for these technologies, and one question that came to mind was are there other uses for the mobile MR technology currently being explored other than imaging the brain, for example, imaging the knee, or some organ that's maybe less complex? Is there research being done and how is that research being carried out? Is it similar to the kinds you've been undertaking? [SHEN]: Yeah I can say just a brief word on that, so the answer is, yes, there are, and — caveat on this: it's not my expertise, but I do track it a bit. In fact I just was tracking a um mr imaging company that is promoting a technology that will allow for assessment of body health in particular, a lot of things around fat in the body and other thing. So this is happening and their marketing plan, this is one example but I think it's illustrative of ideas, is that it will aid clinicians and patients in decision making, and so without naming the company by name, the idea was that they I think they even said like color coded, easy to understand output for you to then guide your patient about she or he or they should do so absolutely this is happening, and MR — Damien may talk about this more broadly — is used robustly on other other body parts; in fact most of us if we've had an MR been like on the knee or you know something something like that so yeah that's that's happening as well.
[SIMON]: Damien, I don't know if you had a comment. [FAIR]: Yeah, no, I was just going to say that a lot of the early technologies on this front have been driven a lot around the brain, but it's coming to review, and it's not just the MR technology that there's a lot of heavy investments in — it's the the associated infrastructure. Once it's collected where did the imaging go? Like there's these are massive in size and there's little tiny hard drives and machines that go somewhere but then how do you get it there and then; is it protected you know because this is all protected health information? So there are several companies — lots of investment around the services about you know storing, grabbing, holding, viewing, analyzing, processing like that service part of it is also being heavily invested in as well, on all fronts. [SIMON]: Again going on this issue of market access how do you see standard setting organizations and stakeholders coming together or working together to ensure there are parameters that everyone can agree on and follow to make sure the imaging we get is of good quality and will be used for the right purposes? [SHEN]: Damien, do you want to say something? I have lots of thoughts from that but yeah — [FAIR]: Obviously I'll let you start because that's the — [SHEN]: Okay, sure, yeah, so I guess several things today, David — it's a great question. So one thing that just has to happen is some of the research, and this is starting to happen, and so you know a basic type of research is how does using the low field scanner, one of the portable scanners, line up to using the traditional fix scanner, and it's a pretty straightforward set of research studies that you run, and these are happening across some hospitals and more many more of these will happen; when the relevant societies the neurologist radiologist will come to a conclusion about whether or not they can convincingly or credibly do the same things with the new technology that they could do with the old, and they'll determine we can use it for this but not for that. I think that has — it takes time
and it's complicated it's not easy but at least it's regularly done when any new technology shows up. I think that unique to brain imaging will be — maybe it's involved because — let me rephrase that one of the things that is especially important in brain imaging and probably also especially important in a few other places like genetics is that the nature of the data collected has implications that can run quite deep because I mean try to save it right the brain is the organ that puts it all together that really defines who we are and so scanning the brain in my view is different than scanning the knee and different than scanning the heart not on the the technical side and for the heart you'd have to also figure out you know under the knee does this new device do the same thing as traditional mr does? For the brain I think you have to think carefully about what would an influx of brain imaging data do and especially do in communities and amongst populations that have never had this data before and with clinician populations that have never utilized this sort of data before and on that front I think the sort of standard setting is more complicated because you have more stakeholders you have more actors, actors who you know haven't maybe done this before, and that's you know part what we're trying to do on the grant is this really diverse set of professions and stakeholders, and I think if I was going to start anywhere however I'd start with the standards around who gets to use the technology and what are the requirements, the training requirements whether you're a graduate student or you're an undergrad, right? This is the technology that undergrad psychology department could purchase or even a high school could potentially purpose. Indeed the grant that Damien and I would like to fund is we would like to take this into middle schools because it could be a wonderful teaching device but we wouldn't do that at the part of the grant is to think about okay how would you train, what does someone have to know in order to use this technology? Both about how to use it, like which button to press, but much more importantly about what to do like Damien just saying about understanding how data is processed understanding what an image is which a graphical creation of the statistical maps, so there's a lot of that work has to be done, and I hope that we're you know a part of making it happen.
[SIMON]: Okay I wanted to shift a little bit based on something you said about data quality this came up in I think both of your talks, specifically the role of AI and machine learning in running these technologies and you mentioned that — Dr Shen, you mentioned that it's important to have a representative population from which to draw data because otherwise you can get data sets that don't track the characteristics in the relevant populations and this was something that happened in the in the genetic testing context, and there was a question about this in the Q&A also that related to, well, how do you figure out which population maybe to test against even if you get all the representative data, and if we start testing based on ancestry, are we moving towards or away or somewhere different from the kind of race-based medicine that at one time was a popular way to — or at least one way to to figure out what treatment to use? [FAIR]: That is a very complex question of course. The call to you know an entirely abandon race from medical research endeavors started several decades ago. In fact the AMA, I believe just last year just recognized race as social non-biological construct for the first for the first time and really you know the way that I've been thinking that I've been thinking about this quite a bit is making sure that how you use race or really what which you oftentimes is used as a surrogate for what you really should be measuring is like some of the structural inequities that exist in our society with regard to socioeconomic status and things like that which is like the the combination of which is kind of being read out as a race thing even though it's not really race it's not really the biology — that you have to be really careful about what your what the question what you're actually asking what are you trying to figure out, because if you're trying to figure out concepts or answers related to the social inequities then races is something you might want to look at, but if you're actually you're trying to understand or develop develop new therapies that are based on the biology of the brain or the knee then it's probably that's when you then you're likely going backwards. I think that the, in your terms, I think there is like a slow movement to kind of recognize what this what this actual difference is, but it's still extremely complex, and there's several new um papers and views out that are really quite good — I should look them up and maybe put them in chat, more recently, describing a more detailed way to think about these this um these particular issues in our research and how to move forward making sure that when we're developing new therapies that utilize these types of technologies that we're inclusive enough such that you're not biasing anything to even potentially be harmful for one group in our society versus another. [SHEN]: If i could um piggyback on that Damien, and I realized we should talk about this because I think there are some really concrete questions around this issue that are just now emerging because of big data and more diverse data sets, and I'll give you one of them. I'm presuming, in most of the data sets, Damien, that you and your colleagues are working in fact I know; NIH requires data sharing at the end. Once you've done it you put up your data set, and there's
going to be you know for each participant all sorts of information and you probably include age; I don't know if you include age the participant in the shared data, maybe you do, maybe you don't — [FAIR]: You can't you include birthday but you can include the age — [SHEN]: Okay, which is I presume because it's thought to be relevant; you don't include hair color hair color completely irrelevant I presume, but you include age and some other things maybe like height, weight, like for some of the developmental stuff for others — who knows whatever you can connect I'm considering. Should that data set include, if you had it, you could have a measure of participant self-reported race? Do you include gender in the data set? [FAIR]: In fact most of the demographic tables there is some ethnicity and race is actually collected. [SHEN]: It's collected but okay — [FAIR]: And shared. [SHEN]: So as that gets shared more and more you got these big data sets, what if someone comes along, not you, someone comes along and starts doing some of these studies that would seem to take us back and draws inferences that you can't do that but like the media really — so it's like a really I think — [FAIR]: That is not theoretical; that is happening today you know, so like the ABC. This is just
happening in the ABCD study where you know this is a big national sample of 10000 people, and we've had to develop groups to kind of read back some of some of exactly what you're talking about you know misuse of some of the information — [SHEN]: Right so imagine that on a grand scale where you now have even more massive imaging, and you're doing it at home, and you've got you know the do-it-yourself — it's a great question because on one hand you want more inclusive data sets for lots of reasons as Damien mentioned, on the other hand you have to think carefully about what does that mean for the practice of research and then clinical care and clinical use, and to my mind we're in a moment of flux, in some ways a good reflection that people are talking about it meaningfully and carefully, but I will say is I think many people on this call know there is — and this is not unique to brain science there are other areas of science as well — but there is a very sordid history of brain data and race in the United States; I mean just just horrible, and I'm always concerned that we replicate that inadvertently. [SIMON]: There's also, just about that point, gender-based misuse of data as well dating back a long time. I wanted to ask a question about — this is more of a technical question but involves kind of legal and policy questions as well also about AI, so assume that we get a representative data set and we start running a machine learning algorithm, some kind of AI ,and what kind of tools do we have to double check that after we've run the algorithm for a year or two that it's still producing accurate data. Do we have to update data sets or if it's using data as it comes in, can that potentially bias the algorithm if it's getting fed data from different kinds of people? Those those kinds of questions is more of like how do we know that it's functioning properly, what kind of tools do we need to use to make sure these mobile MRIs continue to operate accurately that we maybe didn't need in the traditional setting? [FAIR]: That's another really great question. There are — usually in these models that are
being built there's often kind of the user which tries to monitor success and changes and biases and things like that, but as we've seen particularly when we start getting this smoothness of the commercial space, that there's all sorts of conflicts and things you have to you have to consider. One of the big pushes in today's world is related to Francis's comments earlier about how we share data is giving access an open in an open science framework such that there have lots of eyeballs these are the data that's used to develop some of these algorithms and that allow people to test the veracity of them outside of the proprietary user? So I think that you know like further infrastructure to maximize the access to the information and the utilization of that is something that can help avoid some misuses and drifting of some of these models to things that we don't want. [SHEN]: Yeah, and I would just add that you know the systems can operate remotely because the data flies back to mission control where the proprietary AI analyzes and spits out then the image that shows up on whether it's iPad or phone or or the remote location, and I think that transparency will be important, but I think that transparency is going to be in tension with IP and protection of intellectual property; you've just invested and your investors and your company have just given you millions hundred millions of dollars, you can't make that open source, so that's not unique to brain imaging. That's where brain imaging will — we can learn lessons from other areas of law, but we haven't had to confront that in brain imaging before and just as there was ask for a chat, there are many different pieces are neuroscience and race, but there's great new work by Oliver Rollins. I'm putting a one link to an article and then I'll put his book link as well, great, great book [SIMON]: Great, thanks. So we have about four minutes left and usually we like to finish about a minute or two early, so what I'd like to ask each of you to do is comment on the following question or questions, which is relating to this project what keeps you up at night when thinking about mobile brain imaging, both from you know the positive perspective and also from the perspective maybe having some concerns that make you worry.
[SHEN]: Well, I can go first. The thing that keeps me up at night is the advent of snake oil, the sales people um who show up and are going to start direct advertising and running late night commercials and be on radio and are going to snicker people into thinking that this imaging is giving them information that it's not. I think that's, and then taking a whole bunch of money and becoming rich while doing it, I think that's a real big problem and very possible. What I'm really excited about is that this is an amazing technology; it's only getting better; it's not perfect; it's not a substitute for lots of things, but it can be a real contributor to our advancing understanding and then um of significant brain health concerns and improving brain health and improving mental health along the way, so I think there's tremendous potential here, and I hope we avoid the perils. [FAIR]: I think I just have to echo that by far the thing you worry about the most is folks trying to apply these new technologies to conditions and things they can't actually assist with or answer. I mean it already happens. I mean even without this widespread accessibility and with it it's a
it's a big fear that we've got to — we definitely want to be you know ahead of the head of the the game here to assist with avoiding some of the the pitfalls of that accessibility as well, but the possibilities are really amazingly high at this stage in the game and like I was saying earlier you know MRI, it hasn't been around in the scheme of things it hasn't been around that long; we recognize it for things that we use for particularly in the brain you know stroke and for tumors and things like that but really the space and the potential for it to expand into into um functions that don't necessarily have a structural signature is just enormous, and if you think today's world about of understanding and characterizing really complex disorders involved in all of mental health and various neurologic conditions, even in the even in you know in musculoskeletal conditions and you know it's just the the potential is extremely high and I think that over the last you know several decades that we're kind of finally there to have the right context and all the special sauce of the technology, funding, the understanding of all the things you did wrong, all that's kind of coming together all at one point so the potential the next decade I think will be something on this front that we that we certainly haven't seen in the past; the potential is just way high so that part is extremely. extremely exciting. [SIMON]: Well, great, thank you, thank you both for really interesting presentations and a lively Q&A and we hope that everyone has enjoyed this webcast. Thank you for joining us and we hope to see you next time.
2022-02-07