Medical Sphere

🩺 We’re excited to share our most recent collaboration, published at @COLM_conf 2025, the result of joint work between outstanding researchers at the @UW, @CarnegieMellon, @LTIatCMU, @allen_ai, and @CPHAI_Dartmouth, in partnership with @LavitaAI. 📄 Link to paper: arxiv.org/abs/2502.14860

We created a thread for this ECG with all the follow-up questions and tested a group of frontier AI models to see how they responded. Here’s how they did: Most models agreed the ECG shows **sinus rhythm with second-degree AV block type I (Mobitz I/Wenckebach)**, based on progressive PR prolongation with grouped beating and a dropped QRS. The main differences were that one model called it **sinus tachycardia + Mobitz I**, while another interpreted it as **PACs in bigeminy** rather than AV block; a rate estimate of about **78 bpm** also appeared, supporting non-tachycardic sinus rhythm. 🔗 Full thread: medicalsphere.ai/cases/28b1ced4…

From the archives What is the rhythm? Poll next 👇 Digitized using @PMcardioApp

@cingolani_oscar medicalsphere.ai/share/af599390…

@cingolani_oscar Which AI model did better? 🤔⚔️

What’s your diagnosis?

@sara_salehi_MD 👏

Excited to share that our abstract, “Are Pretrained Transformer Models Practical Under Real-World Resource Constraints? A Comparison With Locally Trained nnU-Net for Glioblastoma MRI Segmentation,” has been selected as a finalist for the Miami Minute Presentation Trainee Innovation Award at #IME2026!🚀 Huge kudos to my colleagues @YaldaNikanpour @yashbir143 and my amazing mentor @Slowvak, this wouldn’t have been possible without your support! Sincere thanks to @umiamimedicine for this incredible opportunity! #IME2026 #MedEd #AIinMedicine #Neuroradiology #NeuroImaging

@ASNKidney @AskMedSphere answer this question

Can you answer this case-based, board-style question and apply your clinical reasoning? Check your answer: asn.kdny.info/K9GY50YAW2i Prepare for the boards with BRCU, featuring 450+ case-based questions like this, board simulation mode, flashcards, and more. Register now: asn.kdny.info/k1EK50YAW2j

We gave this image to several AI models and asked: can they reliably distinguish infiltrative non-enhancing glioma from vasogenic edema? Here’s how they responded: All models agree that the MRI alone cannot reliably distinguish a nonenhancing infiltrative glioma from vasogenic edema, though the expansile “butterfly”/corpus callosum-crossing pattern is more suggestive of infiltrative glioma than simple edema. 🔗 medicalsphere.ai/cases/e6614876…

I contend that it is not possible to reliably distinguish infiltrative non-enhancing glioma from surrounding vasogenic edema on standard MRI. Huge literature on this topic. No consensus. Role for advanced imaging and AI? Maybe, but large scale tissue validation is needed.

@BraydonDymm How did these models do?! 🤔 Most models agree the image shows an acute bilateral medial medullary infarction (“heart appearance” sign on DWI), likely in the anterior spinal/vertebral artery territory, which can mimic peripheral weakness/GBS. 🔗 medicalsphere.ai/cases/df5da371…

Consult for "peripheral weakness" MRI shows this. Diagnosis?

@Medzonetv All models agree the spot diagnosis is a **left adrenal pheochromocytoma** (catecholamine-secreting adrenal tumor), based on the **refractory hypertension, episodic headaches, and adrenal/suprarenal mass** on CT. 🔗 medicalsphere.ai/cases/eb7811c2…

History of treatment refractory hypertension + Episodic Headaches. Spot Diagnosis????

@ManualOMedicine x.com/MedicalSphereA…

A 55 y/o man involved in a motor vehicle accident complains of headache, his GCS is 14/15. - What’s the diagnosis ? #medtwitter #foamed #foamrad #radiology

Turn any post on X into a medical AI evaluation. Drop it into Medical Sphere → compare AI responses and vote 🩺👇

@vicky__lou @SocietyofFetMed Haha, thanks for your comment. To be fair, these responses are only from the Gemini models that currently support video input in their APIs. We’ll be adding more models with video input support soon.

@MedicalSphereAI @SocietyofFetMed AI not yet replacing MFM then! 🥴

Question 218: What is your diagnosis? Case Credit: Dr. Alok Varshney (SFM) Follow for more #maternalmedicine #fetal #fetalmedicine #societyoffetalmedicine #fmf #fetomaternalmedicine #sfm #isuog #diagnosis #medtwitter #fetalmed #ultrasound #Radiology

@Rheumatology 🔗 medicalsphere.ai/cases/64fe1990…

Both Gemini models we tested agree this is an **anterior shoulder ultrasound** showing normal anatomic structures centered on the **deltoid/subdeltoid bursa, long head of the biceps tendon, subscapularis/rotator cuff, and humeral head**—with no clear pathologic diagnosis stated. The main differences are in labeling: one emphasizes the **rotator interval/coracohumeral ligament/anterior capsule**, while the other instead identifies **subscapularis muscle, anterior labrum, supraspinatus, and greater tuberosity**.

Uʟᴛʀᴀsᴏᴜɴᴅ Qᴜɪᴢ. Lᴇᴠᴇʟ: ᴀɴʏᴛʜɪɴɢ ʙᴜᴛ ᴇᴀsʏ Aɴᴛᴇʀɪᴏʀ sʜᴏᴜʟᴅᴇʀ. Aʀᴍ sʟɪɢʜᴛʟʏ ɪɴᴛᴇʀɴᴀʟʟʏ ʀᴏᴛᴀᴛᴇᴅ ᴡɪᴛʜ sʟɪɢʜᴛ ʀᴇᴛʀᴏᴠᴇʀsɪᴏɴ. Wʜᴏ ᴄᴀɴ ɪᴅᴇɴᴛɪғʏ ᴀʟʟ ᴛʜᴇ sᴛʀᴜᴄᴛᴜʀᴇs ᴄᴏʀʀᴇᴄᴛʟʏ?

@mandeep_mayo 🔗 medicalsphere.ai/cases/eec3d215…

CT FFR of LAD: 0.69. Do you believe it and would you stent the LAD?

All the Gemini models agree that a CT FFR of 0.69 is abnormal and potentially ischemia-producing, but that management should be confirmed with invasive physiology (FFR/iFR, ideally with pullback) before committing to PCI. The main disagreement is anatomical/therapeutic: one model sees a clear focal high-grade mid-LAD lesion likely to warrant stenting, while another sees diffuse LAD disease with no good focal target and advises against stenting; the others are more cautious/middle-ground.

@dramartaguillen @ClinicaCEMTRO 🔗 medicalsphere.ai/cases/70addbbc…

@dramartaguillen @ClinicaCEMTRO All AI models we tested agree the images show **hand/wrist orthopedic internal fixation (osteosynthesis)**, most clearly a **volar locking plate for a distal radius fracture**, plus **plate/screw fixation of a metacarpal fracture** and another **carpal implant**.

🔩¿Por qué usamos placas, tornillos y clavos? 🥼Permiten mantener el hueso en su posición, favorecer su consolidación y recuperar función, especialmente en mano y miembro superior, donde cada milímetro cuenta Cada fractura requiere un tratamiento individualizado @ClinicaCEMTRO

@SNISinfo 🔗 medicalsphere.ai/cases/dd49e898…

There is no clear consensus across the models we tested on the exact artery or diagnosis: suggestions include the **superior thyroid artery with goiter/Graves**, **ascending pharyngeal or caroticotympanic artery with paraganglioma**, and **ophthalmic artery with cavernous sinus dAVF**.

What artery is marked by the red arrow? What pathology is the most common cause for this artery to be enlarged, as in this case?