Michael Gorin, M.D.

@VickersBiostats I thought this insight was spot on. Do you have others to share? One I have is that AI allows for an endless number of comparisons. Interpretation of p must be viewed in the context of # of hypotheses tested. Detailed method reporting = accurate interpreting of data by readers.

@VickersBiostats I recently had the pleasure of being on a email thread where you shared that in AI res folks report recall and F1 score. In med lit we use terms sens. and PPV. In this new era where AI tools are commonly reported in clinical journals, authors should avoid AI jargon if avoiadble.

Doing statistics well is hard. Avoiding egregious errors in analysis, reporting and interpretation isn't. Just read, understand and follow the guidelines. pubmed.ncbi.nlm.nih.gov/32451178/ pubmed.ncbi.nlm.nih.gov/37286459/ pubmed.ncbi.nlm.nih.gov/40914655/ pubmed.ncbi.nlm.nih.gov/30580902/

@VickersBiostats @ReisLO In this new era of AI, any thoughts on an updated guideline document?

@thekaransinghal @shamay___ Turns or no turns, it should be able to tell a patient with ketoacidosis to go to the ED. All the info was in the prompts in our study. This is absurd. These are people's lives. First do no harm.

Thanks Shamay, totally agree. HealthBench and the independent evaluations I shared (Stanford Med-HELM, MedMarks) also include plenty of single-turn evaluations. I think the key difference is whether we're talking about an LLM generating a natural / open-ended response or answering with a multiple-choice answer / closed-form response. The former is most realistic for evaluation of a conversational system (users don't usually ask multiple choice questions), the latter can still be useful for evaluation of API use cases if done well. openai.com/index/healthbe…

@thekaransinghal When interacting with real patients we dont get to choose how they respond. Whether they are a good historian (your approach?) or a bad historian (our approach?) I figure it out. A consumer health tool should get it right no matter how the info is presented.

What is the reliability and practicability of FAP-RADS Version 1.0 for structured reporting of FAPI imaging in pulmonary and urothelial carcinoma? ow.ly/SFKW50YmVxJ #NuclearMedicine #MolcularImaging #PETscan @ProfKHerrmann @michael_gorin @radnucmp

>10 years ago, @allaf_mo, @somjavadi, @michael_gorin, and I started on a quest to use sestamibi for renal mass characterization. I'm proud that we produced the following story: (TL;DR: it works!) pubmed.ncbi.nlm.nih.gov/41433425/. @AlirezaAmin10 @liljasolnes @nirmishsingla @gottsled

🚨🤖 How safe is AI triage at consumer scale? A structured stress test of ChatGPT Health (launched Jan 2026) evaluated triage accuracy using 60 clinician-authored cases across 21 domains (960 total responses). 📊 What emerged: Performance followed an inverted U-shape — most dangerous failures occurred at clinical extremes. ⚠️ Emergency cases: • 52% of gold-standard emergencies were under-triaged • Examples: diabetic ketoacidosis & impending respiratory failure → advised 24–48 hr follow-up instead of ED • Classical emergencies (e.g., stroke, anaphylaxis) were correctly triaged 🟡 Non-urgent cases: • 35% showed inappropriate escalation 🧠 Cognitive bias vulnerability: When family/friends minimized symptoms (anchoring bias), recommendations shifted significantly (OR 11.7) — mostly toward less urgent care. 💬 Crisis safeguards: Suicide-related prompts triggered crisis messaging inconsistently — sometimes more often when no specific method was described. 📉 Equity findings: No statistically significant differences by race or gender, though CIs could not exclude meaningful disparities. 🔎 Bottom line: Missed high-risk emergencies + inconsistent crisis activation raise safety concerns. Prospective validation is essential before large-scale consumer deployment of AI-driven triage systems. #AIinHealthcare #DigitalHealth #PatientSafety #ClinicalAI #HealthTech #MedicalEthics nature.com/articles/s4159…

🆕 @NatureMedicine How does ChatGPT Health do for appropriately triaging a person as to whether to go to the emergency room or stay home? nature.com/articles/s4159… Not very well. Under-triaged 52% of case vignettes that are considered gold-standard emergencies, like diabetic ketoacidosis or impending respiratory failure

Weekend print for daughters birthday party.

Nice summary of Artificial Intelligence >> Machine Learning >> Deep Learning from @michael_gorin to kick off @BJUIjournal session at #AUA25 Will image segmentation CNNs provide answer to IDing "potentially lethal" prostate cancer (on MRI or histology)? I think so! #SPACE_Study

.@michael_gorin, MD, @IcahnMountSinai, shares the advantages of the transperineal approach over the traditional transrectal method, particularly in minimizing infectious complications caused by rectal mucosa puncture. #Prostate #Biopsy #Urology

Go BJUI Compass! @michael_gorin

@varunshenoy_ @ollama Any update on Ollama integration?

6/ We're releasing Super JSON Mode as a framework with integrations for OpenAI, Transformers, and vLLM. All you need to do 👇 pip install super-json-mode Takes < 10 lines of code to use. Example notebooks are included in the repo. More integrations (like @ollama) coming soon!

Introducing 𝗦𝘂𝗽𝗲𝗿 𝗝𝗦𝗢𝗡 𝗠𝗼𝗱𝗲, a framework for low latency structured output generation from LLMs. Generate JSON up to 𝟮𝟬𝘅 𝗳𝗮𝘀𝘁𝗲𝗿 from OpenAI and open source models. ❌ No need to threaten the model, tip the AI, etc ❌ Built with @derhacobian 🔧 🧵👇

Check out these amazing images of #cinematicrendering in #PETCT. bjui-journals.onlinelibrary.wiley.com/doi/10.1002/bc… #PSMA @michael_gorin @ctisus @Siemens @UNCRadiology @UNCurology

@NM_Urology @EdwardSchaeffer @allaf_mo @DrJennMD @HopkinsMedicine @NorthwesternMed @NUFeinbergMed @McGawGME @LurieCancer @NM_Andrology @EndourologyNm Amazing!! Congrats Dr. Schaeffer!

Our Department Chair, @EdwardSchaeffer, MD, PhD, traveled back to his roots as he is this weeks #VisitingProfessor at John Hopkins Medicine! Dr. Schaeffer completed his residency from John Hopkins in 2007 and is pictured with the current Department Chair, Mohammad Allaf, MD, as well as our newest faculty member, Jennifer Miles-Thomas, MD, URPS, MBA.

@NM_Urology @NorthwesternMed @NUFeinbergMed @FeinbergODI @McGawGME @HopkinsMedicine @MIT @AmerUrological @urogynsociety @SWIUorg @sufuorg Congrats JMT!

We are delighted to announce that Jennifer Miles-Thomas, MD, URPS, MBA, will be joining Northwestern Medicine Urology. Dr. Miles-Thomas will take on the roles of Vice Chair - Regional Integration and Innovation, and Assistant Professor of Urology. With a focus on female reconstructive surgery, pelvic prolapse, incontinence, and neurourology, she brings a wealth of expertise to our team. Dr. Miles-Thomas previously served as the President and CEO of Urology of Virginia. She is currently the Treasurer Elect of the American Urological Association and a member of the American Urogynecologic Society, Society of Women in Urology, and the Society of Urodynamics, Female Pelvic Medicine and Urogenital Reconstruction. Dr. Miles-Thomas earned her Doctor of Medicine from Northwestern University Feinberg School of Medicine. She completed her urology residency and female urology fellowship at Johns Hopkins University. Additionally, she holds a Master of Business Administration and a certificate in Entrepreneurship and Innovation from the Massachusetts Institute of Technology. We look forward to the contributions Dr. Miles-Thomas will make to our team and the positive impact she will have on our patients and community.

@ollama @huggingface I am a clinical researcher @IcahnMountSinai working with @ThomasFuchsAI and @gabricampanella to develop LLM tools for data extraction from unstructured data to be fed into other multimodal models they have developed for cancer outcome prediction.

@michael_gorin @huggingface 🫡 so awesome! What model(s) are you planning to run, and if you don’t mind us asking if you have a specific use case in mind already? Let us all know how we can help! ❤️❤️❤️ Thank you for supporting open source

My newest baby has arrived! ⁦@ollama⁩ ⁦@huggingface⁩