Deepak RJ

MD | Medical AI Education & Research

fan of complex systems, translational AI in medicine, psych

writing cool stuff here:

[Deepak RJ @dyad_rj]

@samarthrawal great work! I've seen @remnote has a "Near misses" section in their flashcards which is essentially the "context" from Pearls but explores adjacent concepts.

e.g. card about metformin could compare/contrast other meds / give counterfactuals of other cases

personally a fan!

[kasey @kaseyklimes]

important essay — one that I hope earns more attention to a horribly debilitating, deeply misunderstood, and grossly underfunded disease

[Deepak RJ @dyad_rj]

Interesting paper that could expand workflows in medical reasoning: https://t.co/37oMtcNavm

The core insight: image segmentation can now 'think' with actual image tokens, so theoretically oncologists could ask 'find me all lymph nodes that look similar to <this concerning one>'

[Deepak RJ @dyad_rj]

@samarthrawal I've found the same! I like to present the case back, especially if it's complex with lots on the problems list and occasionally I find some ways I can optimise care

[geoff @GeoffreyHuntley]

AI is like a guitar. Everyone knows what a guitar is and understands time must be spent to get good at it but there seems to be a disconnect here rn with how SWE are approaching it.

Rick gets it.

[Deepak RJ @dyad_rj]

imagine the impact of Google Beam on telehealth

[Deepak RJ @dyad_rj]

@iScienceLuvr would love to join! clinician/researcher + got my start in medical AI with https://t.co/rAPVf4u41K!

[Deepak RJ @dyad_rj]

4) The coolest foothold I've found lately is @drmichaellevin's work on bioelectricity, applicable to many diseases. These footholds are the highest yield contributions to biomedicine. Therapeutics is just a leverageable foothold. Diseases like delirium desperately need footholds.

[Deepak RJ @dyad_rj]

Just read Wolfram's Computational Medicine (https://t.co/vjuelHNW0v):

Physics has computational irreducibility but pockets of reducibility at various scales allowing for laws (e.g. gas particle collision is unpredictable and chaotic but macroscopically, ideal gas law emerges) 🧵

[Deepak RJ @dyad_rj]

3) Ideally, you want to develop tools to probe disease at different levels (there is no one computational level due to the complex multi-scale dynamical nature of disease) - consider this like identifying footholds on a mountain you're climbing.

[Deepak RJ @dyad_rj]

lol #DeepSeekR1 solves math problems for fun

[Deepak RJ @dyad_rj]

@algekalipso Fascinating thoughts! Just discovering your work. Quick unrelated q - any insights on measuring/modeling attention? It's understudied yet crucial in many neuropsych conditions. Would love any reading recommendations you find especially compelling

[Deepak RJ @dyad_rj]

@owl_posting lol @owl_posting knows who @owl_posting is

[Deepak RJ @dyad_rj]

@NTFabiano @dcperi looks like there are some for late cortical blindness but none for congenital/early cortical blindness!
https://t.co/UHLGhYUP4u

[Deepak RJ @dyad_rj]

🧵3) We then realised there's still 30% of Cambodian women with unexplained anaemia.

TLDR; we have no idea why millions have anaemia

https://t.co/kDZzWmQn6V

[Deepak RJ @dyad_rj]

🧵1) ~50% of children & women in Cambodia have anaemia. We added iron fishes to help improve it (assuming it was iron deficiency).

[Deepak RJ @dyad_rj]

🧵2) turns out we just ended up being iron overloaded - it was not iron deficiency. We then realised it was likely thalassaemia that we'd accidentally tried to fix with iron fishes.

https://t.co/8PnqC4wLAE

[Deepak RJ @dyad_rj]

STL files: https://t.co/PkmlmIYk6z

[Deepak RJ @dyad_rj]

$9 | 3D-print your microscope
$100 | add a camera + LED + raspberry pi
$200 | diagnose malaria???

https://t.co/UpSwZFopVN