Good morning.

Good Morning from Dallas TX

We will be muting everyone but if you are not muted, please do so if you are not speaking.

Good morning from San Diego! Good to see you all again!

Good morning everyone!

Welcome! Please mute your mic if you are not presenting.

Good afternoon from Northwest England. Thank you for all you massive contribution in recent times.

Good morning

Good afternoon from Nairobi. Great to be here

https://hbr.org/2021/07/ai-doesnt-have-to-be-too-complicated-or-expensive-for-your-business?utm_medium=email&utm_source=newsletter_daily&utm_campaign=dailyalert_actsubs&utm_content=signinnudge&deliveryName=DM143580

> “All of AI, not just healthcare, has a proof-of-concept-to-production gap,” - Andrew Ng> "There are challenges in making a research paper into something useful in a clinical setting" - Andrew Ng

Thanks Arlen! Good morning everyone from Charlotte, NC!

Intelligence-Based Medicine is an excellent read. Well worth the time.

And, how do we regulate AI? https://flipboard.com/topic/opinion/what-the-u-s-can-learn-from-europe-s-approach-to-artificial-intelligence/a-XGPLonWbQxiAuK8ec1MJ4Q%3Aa%3A3195393-9bae2ea0ef%2Fnytimes.com

ABAIM Course Review & Testimonial - https://youtu.be/ho29j_3_2nE - Must attend

If there are any questions, please post to the chat. At the end, we will also have an opportunity to ask questions live.

A very warm welcome to everyone from Southern California! Hoping to see you in person next summer!

MIS50 is the special 50% discount for our summer summit until next Friday, August 6th.

AI-MED.io

See you next Friday at ABAIM/SOPE office hours at www.abaim.org, 8am MT. Subject: AI and cybersecurity

Good morning (Laguna Beach, CA)

GPT 3 is by OpenAI

Another advance: Healthcare AIntrepreneurship

Please post any questions for Dr. Nguyen

How should AI be regulated?

Dr Nguyen re: your slide on federated learning surely there are a lot of other institutions as well as hospitals accumulating and using data?

What is the future of blockchain and AI?

Can we leverage this swarm AI with disparate levels of data quality and non-normalized data?

Each medical society in all medical fields and subspecialties can have a task force to collect data in their fields and come up with a guideline platform to be used by other doctors, particularly in rural area where there are limited resources.

Totally agree with Panida!

I will let you know as I will be working on this. There will need to be some data quality surveillance.

We all should spread the words to all medical societies we belong to.

Can we get a copy of the Slide or a PDF of Bibliography? Thanks! This is great thanks @Nguyen

We need a better design of VR/AR headset to reduce motion sickness when using it.

@Panida yes

AI and blockchain will be essential parts of swarm learning

XR + AI = Intelligent reality so clinicians can train without a full team or a mentor especially for resuscitation training

XR, AR, VR will be very helpful in treating Psych patients. They need multiple touch points besides seeing the clinician once or twice a month! EE

Do you see eye tracking built in to XR devices, and are we using the data generated from eye tracking to understand how people direct their attention?

First Application of XR would be education and wellness training. Ambient intelligence with virtual assistant does not really need XR/VAR

Eric’s comment on using XR for mental health is spot on. Mental health issues loom large now with the pandemic.

Eye movement as a mode for communication as well

Yes, mental health is very important. Psychotherapy via XR would be nice.

it would be nice to be able to assess healthcare team dynamics, such as an expert observer to help train people to work better together

Yes

How about virtual meeting via XR??

A friendly note to speakers in 15 minute sessions: Please try to complete your talk in 7-8 minutes as to leave at least 5-8 minutes for questions. Thank you.

Thank you Joe for enlightening everyone on how AI can be combined with XR to take the latter into a new dimension!

Congrats to Fatme for using AI as a valuable resource in challenging milieu.

Eric - completely agree re: psych treatment and XR. Things like avatar therapists, CBT, education, and delivering insights can be combined with environments focused on things like relaxation, mild exposures, and the like to bring clinical impact.

A special welcome to our international attendees from all continents!

@joemorgan, thanks for an excellent presentation on XR and the future possibilities. EE

Congratulations to Fatme for bringing an international perspective to Lebanon to include AI

Fatme, how can we facilitate curation of your clinical expertise into future ECG knowledgebase?

@FatmeCharafeddine, I applaud you for continuing the science under such dire conditions in Lebanon for your patients! You are an inspiration to all of us. Praying for quick recovery and healing of all the Lebanese people! EE

Very inspirational!! Congrats!

@Fatme really inspiring!

Well done @Fatme ….. incredible.

Fatme is a super spreader of AI !

@Fatme thank you for doing this - you are an example to us globally as well as representing your country admirably!

Well done Fatme. I've been looking into this issue for a long time now, I think a proper implementation of AI is the only way forward to democratize access to expertise

Thank you Fatme for a wonderful example of perseverance and dedication to helping others

Thanks Fatme, that was inspirational

I am working with Timothy Chou on a swarm learning concept for images but also data in general for children called AI4Kids.

Fantastic talk Fatme! How far away are we from moving beyond 12-lead EKG to EKG vests with many more leads, interpreted by AI? Would it even be helpful to use many more leads?

Inspirational and aspirational. Great work.

excellent talk !

Digital twin talk is by Dr. Hamilton Baker at MUSC

Eric, do you think the failure averse culture of healthcare is an impediment to these efforts?

Eric and his triple perspective is extremely valuable for all of us to appreciate.

The order for telemetry is most often a reflexive decision

Most of hospitals and healthcare systems require a learning requirement in various topics every year, we should add AI in healthcare in this learning requirement to give all the staffs to know about AI.

The MIS50 50% discount code for AIMed Global Summit 2022 in January is now up so please let me know if you are having any issues signing up. This discount will expire by next Friday. Ai-med.io

This variation stems from human biases and data sparse heuristics...mainstays of 20th century mediciine

We have a two day review course under the ABAIM (ABAIM.org) that is very comprehensive and live. Next dates are August 13/14 and there are still a few slots available.

Please put your questions into chat for EE

Great lessons learned from Eric. Thank you for sharing!

If the AI determines outliers and its tied to co-variation incentives. Does it work well? Would you worry, it also encourage group behavior that may not be the best.

Question for Eric, can you elaborate how Amazon is a competitor and where are you seeing their early success in healthcare?

describe tech and digital security re cyber threats contrasting to holistic view across sites?

The fail fast mentality in healthcare seems inappropriate - but with little or no liability or responsibility, it seems like it will continue. The latest issue with EPIC sepsis algo seem to highlight this - liability seems to be outsourced to the practitioners, not the AI provider - and the gag clauses prevent any open discussion to fix things

Would you have a methodology for evaluating vendors for AI solutions. Is there a curated list?

What is your view on keeping patients at home while monitoring them from the hospital? Basically, virtual hospital. This can help reducing a risk of confusion and deconditioning in geriatric patients. So, patients who can be treated at home, they can choose this option.

It sounds like Amazon is taking an omnichannel multi pronged approach to data capture and normalization into AWS

@George Mathew, I think the Epic sepsis issue is really complicated. I agree that the liability is outsourced but I think the main issue is what was touched upon earlier. Epic created the model to be easily shoehorned into all their client sites. I know of some orgs that have been successful with the tool but many that have not. My opinion is that to a large extent the issue is the underlying data, not necessarily the model.

how to handle bias and SDOH?

Great presentation Eric! Thank you for sharing!

@JohnLee I don't disagree - but EPIC is beta testing while its learning this - I don't think the risk is being adequately communicated to patients or to practitioners - agree that the underlying data is not perfect - but there's a lack of intellectual dishonesty in not revealing that imperfection and the downstream implications to all parties - its not perfect, and people need to made aware of all of the risk, not being sold on something that will "work" - even with multiple iterations

sorry meant lack of honesty - apologies for the rant

Can you comment on Amazon's HealthLake - they claim to have ML resources and tools for healthcare data. Have you looked into this? Institutions with limited computing infrastructures may find this useful.

and bias of hospital level characteristics

easy to create league tables of surgeons or hospitals but what safeguards that it's not all spurious variation. are funnel plots being applied?

Jeanne symbolizes how clinicians can work synergistically with data scientists at a high level. Thank you for your contribution Jeanne!

@George Mathew I think we actually need to be ok with "beta testing" along the lines with Eric's fail fast mentality. In my exposure with many Epic orgs, they use the Epic predictive models using a fire & forget strategy. They don't fail fast...they just fail and forget.

Actually the fail and forget mentality is rampant in all healthcare orgs, not just Epic orgs

It would be great that pathology result will be available while the patient is still in the OR so that the surgeon can plan for the next step right then and right there.

We need incentives to align for model and data longevity in practice. Again, clinical impact and not just a model rolling out with little accountability for longer term success. Same with institutions publishing and forgetting to deploy widely (“publish and forget”).

@AChang definitely agree

What Panida has suggested is being done at a few centers already: a CNN model in the OR for surgeons.

Thank you, Dr. Chang, Which centers, and can we adopt the model? Or, it is just at a research level?

Please submit your questions to Dr. Shen

@Jeanne Shen, beyond the image processing, do you see opportunity in separating the wheat from the chaff in the readings? What are the truly important nuggets in different clinical situations and consumers of the reading?

@Jeanneshen do you see the merger of pathology and radiology fields in the future with AI being the conduit? Maybe Radiopathologist? Thx EE

There are so many models and platforms already developed and successfully adopted. But, there is no single place to list all of these models and platforms. Do you think MIS can be the place for all these to be listed? It will make it easier for all of us to look for ideas, without having to search Google, or starting from scratch.

Great talk Jeanne and congrats on pathology really embracing AI now. Perhaps slipstreaming behind radiology but very impressive nevertheless.

@Eric, I think that is a future for cardiology, specifically Electrophysiology and Imaging, my particular interest! Check this article, they improved accessory pathway detection by incorporating a CXR into ECGs https://www.nature.com/articles/s41598-021-87631-y

I think there is a great future for fusion AI modalities

@Zaidon, thank you for sharing this multimodal AI algorithms. So exciting!

Multimodal AI is a big part of future of AI in subspecialties.

Great perspective from Matt about AI in a big medical center with so many stakeholders. Big fan of AIMI and what they have done.

Thanks @zaidon. Really good article. EE

Matt will be part of our ABAIM review course guest faculty in August. Thanks Matt!

Please submit your questions for Matt’s session

what is your/stanford's take on federated learning

please share more on the race/ethnicity when applying predictions? scores and SDOH for example in a 75% Hispanic population.

Special request for all attendees from the Society’s first working group! Please help focus the working group’s efforts on the most clinically important areas by voting on use cases and submitting your own use cases at: https://www.reddit.com/r/GrApHAIUseCaseQueue/ I’ll explain further in my talk, but if you can go there during the breaks, we would very much appreciate it!

stay tuned to our website/social media for new results re: race recognition on medical imaging by AI systems

Really insightful and fascinating talks today! Thank you to all the speakers for preparing the presentations and inspiring all of us to continuing moving AI forward in medicine

Please put your questions to chat! Thx EE

What is the next step for med students when it comes to commercial R/D and launch?

Can we use this cap for a long-term monitoring? This can predict clinical seizure and give a warning to the patient real-time.

What else might be diagnosed or monitored using this cap, besides a seizure disorder?

The problem with psychogenic seizure is that EEG electrode cannot detect signal from medial temporal lobe, or orbitofrontal lobe.

What's it like working with dry leads - ?artefact acceptable, ?are they cheap and accessible or proprietary technology?

When medical students invent things, who owns the IP?

An admirable project and I wish you well in producing a marketable product. What about non-baseball caps for other countries without baseball?

I like the collaboration between medicine and engineering. We do have a lot of challenges with workflows and I think this form of collaboration will result in some solutions.

How can we encourage more that medicine isn't a single person seeing a single patient anymore? Collaborations with data scientists, engineers, etc. is necessary for any of us to keep up with what's the latest in any medical field

There are approximately 200 US allopathic and osteopathic medical schools. A recent non-scientific search indicates that very few schools offer mandatory data science/data literacy education and training to med students or residents.

The same holds true for biomedical innovation and entrepreneurship/business of medicine.

Question for the panel: Medicine continues to make technological leaps and bounds without necessarily improving care in terms of communicating with patients and establishing goals of care. How can the integration of AI improve this deficit?

Educating doctors and patients is but the first step. The holy grain is changing behavior, which is much harder. Education alone does not change behavior

Speak their language (financial literacy) and show administrators the ROI.

will the new medical school model will be longer than typical class?

How much are online learning tools like OnlineMedEd, Coursera, Khan Academy, etc. disrupting the traditional lecture model of didactics in your school?

The new curriculum will be within the same timeframe but different approach and allocation of time.

By democratizing access to expertise. One big issue with medical systems is variability and access. Practitioners in remote areas and developing countries are underpowered and unsupported.

Can we have our own digital twin? Our AI twin will do a non-human part of work and let us free to do the human thing.

And by underpowered I mean access to sophisticated diagnostic systems. If you can diagnose heart failure from an ECG, and we can, how much help would that be to practitioners in underserved communities?

That’s why we need to work all together Zaidon to lead the way in underserved areas, and initiate this revolutionary paradigm shift in culture among the healthcare practitioners and medical students.

Will young trainees lose intuition in human biology and ability to correct machine error, as we introduce more tools for clinical decision support.

Anthony, I needed to hear that! I'm in awe of that perspective, even if I agree with the diagnostic power of a mobile US!

Even stethoscope can incorporate AI

There will be time that you need the old tech.

Humanizing the medicine is very much needed … The stethoscope might not give the diagnostic ability as an ECHO but gives the human touch element and satisfaction to the patient …. !!

As a practicing cardiologist, the stethoscope adds very little information to my clinical routine, and I do round with a portable echo that I take every opportunity to use. But I still use the stethoscope every time because of the human touch, and patients seem to like it

What are the biggest barriers to medical school reform? How to overcome?

Biggest barrier is behaviour and inability to bring about behavioural changes

Nudges maybe

Slow gradual pressure with empathy and patience.

True Dr Chang:)

I would argue that nursing informatics programs are more advanced than medical school informatics

My understanding is that nursing programs in Southern California are fully enrolled with student-wait lists.

Ambient Health by Nuance can be very helpful for documentation while being with the patients.

should medical schools offer alternative pathways for those who are more interested in non-clinical careers?

What have you cut out of traditional medical curricula to make time for engineering and entrepreneurship training?

Accreditation and licensure are challenges to overcome and ultimately adjust to allow the type of education for the healthcare team (physicians, nurses) to evolve (modernize?) and result in the delivery of care that we are discussing.

Tech is wonderful. However, it also dehumanize medical practice as modernizing medicine disassociate the communication of care.

Medical schools will eventually be one of many steps in career development, starting these day in P-undergrad and continuing in residency. consequently, you don't need to teach it all in 4years.

Let's not forget, as physicians it is our responsibility to "extract" relevant actionable information that lead to a diagnosis and a treatment. That information may lie in the history, in the stethoscope finding, or in their "data".

Great panel

Wonderful panel, thank you!

Great comments from our attendees!

Excellent discussion,Thank you all

Eric, great job moderating!

The education should start at the primary school level.

We all use online videos to supplement our learning on in-class. A number of medical school academic affairs administrators at different schools I spoke to understand that online videos may even be the primary tools we use to learn the information. Thus, it is important that any in-class sessions be a more interactive approach that helps build or reinforce knowledge. Our school stresses problem based learning, in which small groups get together to focus on the diagnosing process, which is something that is more difficult to learn from just watching videos.

Thanks for a great panel discussion! EE

I agree, EE! Great session. Another plug from the working group - if you have a few minutes during the break, please go to https://www.reddit.com/r/GrApHAIUseCaseQueue/top/?t=all and vote on use cases or add your own.

BTW, this is an anecdote from last year, forgot to mention it earlier, some may find it interesting regarding pitfalls!Intensivist: Can you please do an urgent echo on this ventilated trauma patient? I hear a murmurEcho fellow: Done. Nothing out of the ordinary.Intensivist: But I hear a murmur, are you sure? Did you listen to the chest?Echo fellow: Why would I listen to the chest when I just did an echo? I don't even carry my stethoscope anymore.Intensivist: Here, take my stethoscope and listen.Echo Fellow: Takes stehoscope from intensivist and listens. There is a continuous murmur, becomes red faced. Repeats the echo with one window that she missed earlier and discovers a PDA (patent ductus arteriosus), a rare congenital abnormaility that can only be uncovered by a certain maneouver (that is part of the routine workflow but ignored by many in the course of POCUS/acute echoes!)

Very good example. We cannot rely on just technologies.

Good story. We should take advantage of both to maximize dividend. Most cardiologists would agree with not having a stethoscope as it does pick up sounds that even good ultrasound can miss, especially screening ones.

interesting playlist

Good story. We should take advantage of both to maximize dividend. Most cardiologists would NOT agree with not having a stethoscope as it does pick up sounds that even good ultrasound can miss, especially screening ones. Trouble with relying solely on one technology is that you are limited with the performance of that technology. Why not use BOTH the stethoscope and ultrasound? BTW, stethoscope is more intimate with patients as you can look at the patients whereas ultrasound distracts you as you will be looking at the images.

My mantra in medicine has always been to adopt the best of the old and the best of the new so I still plan to carry a stethoscope

Welcome back everyone!! We will get started shortly

We are so proud of you Alan and your AI journey and what you do for all of us!

Thanks Anthony and team for your trailblazing leadership and innovation!

Thanks Ji Lin for this amazing work from MIT!

Please feel free to drop questions for Ji Lin in this chat, thanks!

With the microcontroller, does the machine learning algo need to be hard coded?

Very impressive and would help democratize AI

We will have Q&A in 1 minute, please feel free to unmute and ask questions or drop in the chat

awedsome

Awesome!

Tiny but huge

excellent presentation

Can you please share the website here again?

Where do you see the biggest impact of this technology in healthcare?Hospital or outside hospital?

Thank you.Amazing work again! We do see great value in the BP applications

impressive project..

Get ready for the next session starting in 1 minute!

Thank you Ji Lin for this monumental work that will have really big application potential.

That pyramid is WIKID!

Great slides and conceptualization of this domain. Great work Tim!

Please submit questions here or unmute during the Q&A session

Are these diagnoses pulled from ICD-10 codes? The problem with ICD-10 listed in the EHR is that some of the ICD-10 codes are not the final diagnosis, but just a code that put in for sending labs or images. For example, if I have a patient possible Multiple Sclerosis. When I order brain MRI, I have to put in my suspected diagnosis. But, it might not be the final diagnosis.

I think the example that you have described in your presentation for a person who fall from 2-story building is called co-occurences but my question is how do you filter out the irrelevant diagnosis / trauma.

What are the pitfalls of this approach that one needs to be cautious of?

I do not use microphone.

Is the advantage of graph databases with nodes and edges that they are better native visualizations. Is it a more efficient way to connect data?

For the data scientists,preprocessing is important part.You don’t want social or family history getting into problems

Thanks!

Get ready for our next talk from Dr. Abudlhussein in 1 minute!

Great topic👍

This is really imprtant

Thanks again for the opportunity to represent the working group! We would love to have your input on which use cases should receive out attention going forward. Please vote or add your own at https://www.reddit.com/r/GrApHAIUseCaseQueue/

Please drop your questions into the chat or feel free to unmute during the Q&A session

Are there any opportunities for non NHS physicians too?

Great work Hatim that you are doing in the UK and looking forward to collaborate even more. Glad to have you be part of the student/trainee section of the MIS.

Agree with Anthony.Very important work!

The technology also has to be affordable.

Wonderful insights!

How can one be part of such a multifunctional team?

ABAIM review course and Dr Hoyts MIS training opportunities are great.

MIS website lists many learning opportunities

Great. Thanks :)

Thank you Dr Piyush 👍

There will be a 15-minute break now, please return at 12:30pm PST/3:30pm EST for Session 5 moderated by Dr. Hamilton Baker!

Great talk Hatim!

Great talks

Having a great time

https://www.futurelearn.com/courses/artificial-intelligence-in-healthcare

this free course is open to international audiences

and a great starter for ten to understand how we might be applying AI to our healthcare sustem

Thank you Dr Hatim

https://www.misociety.org/

https://abaim.org/

Great job with moderating Alan!

yes

Can see the slides

Send any questions for Quinn for the discussion, thx!

How did you decide which cells to crop?

Very impressive works by our speakers in this session. Thanks for moderating Hamilton!

It does appear that federated learning performs external validation, unlike more models that come from one medical center

Johnson, great talk! EE

Great talk. I can already think about applications in chronic pain management.

This is very helpful for collaborations between academic medical centers. Could be really great also for rural health partnerships to pool data. Great talk, Dr. Qian!

Helpful articles/references

Very good summary talk on federated learning and great idea to consider.

https://arxiv.org/abs/2101.04800

Personalized Federated Deep Learning for Pain Estimation From Face ImagesOgnjen Rudovic, Nicolas Tobis, Sebastian Kaltwang, Björn Schuller, Daniel Rueckert, Jeffrey F. Cohn, Rosalind W. Picard

I am intersting, if any collaborative AI project in the Pediatric Cardiology especially advance imaging. moh100hassan@gmail.com

This is great for traiging!

👏

was delta variant predicted by AI models

AI models have a difficult time with COMPLEX phenomena as these are much less predictable.

Great review of AI tools in COVID by Alan.

Great talk Dr Alan 👍

Great talk and summary!

Thank you for interesting presentations and insights. I am big advocate of patients owning their own data and controlling access to their health data. How would one selectively share health data with a provider or any other party and then at any point in time withdraw or block access to their health data?

Thank you everyone for your time at the MIS Summit! Feel free to connect via LinkedIn - https://www.linkedin.com/in/everyounghealthmd/

Great job by all the speakers in the session!

Thank you Hamilton for moderating!

We will start session 6 in 2 minutes! EE

Please put your questions for Dr. Lei on the chat. Thx EE

Undiagnosed PDAs are a cause of cerebral emboli, known as paradoxical embolization in adults

Yes, I have treated adults as old as 72 years old who had stroke due to PDA and had no idea their whole life they had it. EE

Why not use only RNN for this study since it is dynamic? EE

Are you considering applying this method to identifying other congenital heart defects?

Would you have been able to use the EKG tracing to decide when is diastole, then use those images?

Can the algorithm classify the pda small vs moderate vs large …. That s the actual question neonatologists look for when ordering ECHO for PDA for treatment implications

Do you foresee that the AI consult service will evolve from a purely research service to a clinical service as algorithms are implemented into clinical workflows?

Would you consider spinning this off as a service available to orgs without these resources?

ie: data science as a service

I'd like access to that IPO

@Hamiltonbaker that was an awesome blue print for AI services! Thanks for sharing your insights. EE

@Hamiltonbaker - I recommended that a local university develop a ML/AI consult service 4 years ago and they laughed at me. You are ahead of the power curve

@Roberthoyt - Thanks! too bad they didn’t listen to you. I hope it is a viable long-term model that will be widely adopted