
Radiology resident and BioDesign Innovation Fellow @MontefioreRAD. š° Journal Rounds: new papers, summarized
Joined January 2012
- Tweets158
- Following624
- Followers597
- Likes2.7K
Pinned Tweet
Radiologists are substantially underrepresented among OpenEvidence users, for a variety of reasons. Iām building an alternative combining the best of StatDX, Radiopaedia, and OpenEvidence: an image-rich platform with responses tailored to what radiologists actually want.
Itās currently in a closed alpha with a group of radiologists giving feedback. The goal is to have everything a radiologist would want within reach.
Join the waitlist if you want early access. Feel free to share anything youād want this platform to have (DDx, report error checking, calculators, etc.).
Iām also trying something different with this app. Iāll post periodically about our learnings as we build it. I think itās helpful for physicians to understand how RAG and other LLM powered clinical/education tools are built, since evaluating their outputs takes a slightly different way of thinking.
Hopefully this is entertaining for anyone interested in AI, or for physicians interested in getting involved in this kind of work. We shall see... and wish us luck...
Donāt worry, all images have the appropriate license ;)
Made possible thanks to @claudeai @ClaudeDevs.
openroentgen.com
Join the waitlist
openroentgen.com/
3
218
So are we now defining any non-physical clinical interaction as potentially replaceable work, lumping all sorts of clinical activities and their idiosyncrasies together? Interesting definitionā¦
The authors analyze the degree to which clinical specialties are shielded from, or threatened by, AI depending on how much of their work requires in-person physical interaction. Learn more: nejm.ai/3UDkrIv

ALT Bar chart titled "Limits to AI Substitution across Specialties: Composition of Medicare Part B Payments Including E&M Services." The chart shows percentages of "Physical," "E&M," and "Nonphysical" services across 15 medical specialties and all overall (US$44.9bn). Total 2023 Medicare Part B payments are shown in parentheses (US$ bn); amounts are in billions of U.S. dollars.
Great book for anyone interested in understanding how the American healthcare system came to be, even if you donāt necessarily agree with the premise.
174
Abdullah Norain, MD retweeted
Malpractice case: lumbar spinal mass, $1M verdict.
Take a look at these MR images from a separate teaching case of the same pathology and site as the litigant.
"Contrast-Enhanced T2-Weighted STIR Imaging of Pituitary Adenomas"
doi.org/10.3174/ajnr.A9253
@iantmark @mayoradiology
Abdullah Norain, MD retweeted
I see at least 4 errors
Just asked Claude to do the labelling, and it did it perfectly. Amazing!
I fear our prefrontal cortices will become redundant in the near future š
#MedTwitter
Radiology resident here. I've been experimenting with Jev from @typesafeai and built JevRad, for educational use only.
Describe a patient ā Jev picks the ACR scenario in 0.5s, then selects the most appropriate imaging studies.
jevrad-web-production.up.raiā¦
cc @CompleteSkeptic
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?
Iāve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev
⢠20-200x faster
⢠40-400x cheaper (w/ output tokens free)
⢠Frontier composable intelligence optimized for decisions
AFAICT the shortest path to AI-based economic revolution
āBecause unaided review always preceded AI-assisted review of the same case, recall and reading-order effects may influence the reported benefits.ā
A+ for honesty (buried in the limitations). But if they're going to claim a 50% reduction, they really need a better study design.
Nvidia just released NV-Reason-CT, an open vision language model for CT chest and abdomen analysis. They trained on recordings of radiologists reasoning through cases. In a small study of two radiologists reading 10 scans, using the model output (draft reports essentially) saved them 50% time.
arxiv.org/abs/2609.27511
1
1
563
I wonder how many residency and fellowship programs out there are using LLMs to rank applicants...šš§
1
1
149
Link to the article: pubmed.ncbi.nlm.nih.gov/4272ā¦
55
Abdullah Norain, MD retweeted
What I learned during that time! #vascsurg vascsurg.me/2022/09/10/the-nā¦
A popular request from users: now you can type in your medical specialty and get a curated list of journal recommendationsāno need to type in major journals manually. I hope this can make the journal selection process less painful for folks.
App Store:
apps.apple.com/us/app/journaā¦
236
šØ NEW STUDY: 900k+ patients
5 PM doctor appointments ā
ā 10.6% worse medication adherence
ā 8.8% fewer prescriptions filled
vs 7 AM visits.
Later slots ā rushed visits?
Primary care physicians are under relentless strain ā and the system is failing them
1
1
145
For a while they ran ads proclaiming to be āThe worldās first long term doctorā. The hubris is baffling. This is what happens when investors and founders minimize physiciansā perspective.
Post-COVID we saw plethora of glorified āclinic+appā startups that never made sense.
After $500M+ in funding, 8 yrs in operation, Forward Health is shutting down
The bold vision of high-tech Care Pods aimed to transform primary care but faced challenges with operational complexity, scaling infrastructure and reactive care limitations. Are care pods still future?
2
529
Abdullah Norain, MD retweeted
A wonderful collaboration bringing in local #Bronx high school students. We toured CT, US, mammography and x-ray. Terrific speakers- Dr. Norain #radres, @BeydeNatasha , nursing and Catherine Estrampes. @MontefioreNYC @EinsteinMed #HealthEquity
Our inaugural āWhy Radiologyā STEM event was unbelievably inspiring! In collaboration with @BronxScienceHS and @GEHealthCare we hosted students in Radiology showcasing what we do as physicians, technologists, nurses, PAs, administrators. #HealthEquity @RadHealthEqty š©»āļø
š Clever use of the Jaccard index to quantify quality of residents reports!
#Radiology #ResidencyTraining #AcademicRadiology #MedicalEducation
3
295
Interesting study comparing ChatGPT-4.0 and 3.5 on oncology exam questions. General models will always struggle. Domain specific is the way to go ā we are working on something at @RecapsuleAI. Stay tuned! š
#MedTwitter #Oncology #AI #MedicalEducation #radiology @RadiologyACR


































