
Advanced Cardiac Imaging and Cardio Oncology FIT @YaleMed; Alumni @UconnHealth, @Rutgers_NJMS, Seek truth not confirmation
New Haven, CT
Joined January 2012
- Tweets138
- Following3K
- Followers429
- Likes786
Maciej Tysarowski retweeted
To add some clarity: you don't pay extra. It's the same subscription, same price per month.
What's new our sub now covers two separate pools:
· Interactive → sub limits, unchanged
· Programmatic → new $20–$200 included(!!) credit, metered at API rates
Starting June 15, paid Claude plans can claim a dedicated monthly credit for programmatic usage.
The credit covers usage of:
- Claude Agent SDK
- claude -p
- Claude Code GitHub Actions
- Third-party apps built on the Agent SDK
Readers added context they thought people might want to know
Previously, programmatic usage like claude -p counted toward subsidized subscription limits; starting June 15, it draws from a separate $20–$200 monthly credit metered at full API rates, while interactive limits remain unchanged. support.claude.com/en/articles/15…
Maciej Tysarowski retweeted
From the Kwan Lab @Yale @YaleCardiology to the national stage 🚀Huge congrats to Usmaan on being named a Goldwater Scholar, one of the most competitive science honors in the U.S. Now onto multi-omics + AI to predict cardiotoxicity. tinyurl.com/4mdshsup
Maciej Tysarowski retweeted
in med school i spent a few weeks rotating at a top ICU in Germany
it made me appreciate how strong clinical care is at major academic medical centers in the states
the uncomfortable reality: US healthcare is best in the world if you ignore affordability
Maciej Tysarowski retweeted
“Physicians building with Claude”. A live webinar organized by @AnthropicAI - join us on Thursday April 23rd at 10am PT
Together with @grahamwalker we will showcase our journeys of building for healthcare.
Daisy Sophia Hollman from Claude code team will join us to discuss how to use @claudeai and Claude Code to bring your next big ideas to life - while respecting safety, privacy and compliance.
Thank you @PollyIsrani and Neel Patel for organizing. Looking forward to this great event - register here!
anthropic.com/webinars/claud…
@nosajab @bcherny
Maciej Tysarowski retweeted
It’s 11th year and counting! Teaching the first lecture of @cs231n every year has been a highlight of my spring seasons. As usual, I asked students which departments or schools they come from @Stanford . Increasingly, students raise their hands to indicate that they come from all seven schools on campus, from @StanfordEng to @StanfordMed @StanfordHumSci @StanfordGSB @StanfordLaw @StanfordEd @stanforddoerr . AI is truly a horizontal technology that excites students across all backgrounds and disciplines!🤩
Maciej Tysarowski retweeted
Just reached a major milestone of 200 peer-reviewed publications on Pubmed!
Thanks to all of my mentors who taught me the ins and outs of research and to all my friends and colleagues who supported me along the way
#wilbertaronowMD @ziadalinyc @MartyBLeon @hahn_rt
Maciej Tysarowski retweeted
I just took over as the Program Director of the advanced cardiac imaging fellowship program at the Cleveland Clinic. I’m going to be at ACC 2026 and I will be happy to chat with you about our program! We are currently interviewing for July 2027! Please message me! #ACC26
Maciej Tysarowski retweeted
“Kids I have good news. Daddy is out of Claude tokens until 3PM. He has time to play with you now.”
Maciej Tysarowski retweeted
Welcome to US Cardiology 2036!
beckerscardiology.com/cardio…
~40k physicians for demand of ~47k
(my math: 47k minus 40k = 7,000 shortage)
State-by-state breakdown is huge (MA 147% vs ID 52%)
Where do you practice? Are you considering a move?
Most people think world models are about robots understanding the physical world.
But medicine may be an even better testbed.
First — huge congrats to @ylecun and @amilabs on the ~$1B bet.
Major reason I think world models will play an important role in medicine is that, medical data breaks pixel reconstruction in two fundamental ways.
1. Noise is part of the physics.
Ultrasound speckle depends on probe angle, operator, and anatomy. CT has beam hardening. MRI has motion ghosting. Pathology varies with staining.
Train a model to reconstruct pixels and it spends its capacity explaining scanner noise instead of physiology.
2. Physiology is a process, not a snapshot.
Cardiac function — LVEF, wall motion, valve dynamics — is inherently temporal. A single frame cannot capture it.
A model that predicts forward in latent space learns physiology. A reconstruction model mostly learns appearance.
These are not the same thing.
We tested this directly with EchoJEPA (with @alifmunim):
JEPA trained on 18M echocardiograms from 300K patients.
Results:
→ 20% lower LVEF prediction error vs reconstruction baselines
→ 79% accuracy with 1% labels (baselines: 42% with 100%)
→ 2% degradation under acoustic artifacts (baselines: 17%)
→ Zero-shot pediatric transfer beating all fine-tuned models
A 37-point gap at 1% labels isn’t a hyperparameter story.
It reflects what the representation actually learned: cardiac structure and dynamics — not acquisition noise.
The bigger lesson: there’s a structural mismatch between what reconstruction learns and what medicine actually needs.
Replying to @ylecun
@ylecun's vision: machines should learn like humans — by building internal world models, not reconstructing every pixel.
We just validated this idea at the largest scale ever attempted in cardiac ultrasound.
Introducing EchoJEPA — the first world model for medical video.
🫀 18M echocardiograms
👥 300K patients
🧠 Learns heart dynamics — not imaging noise
The problem:
Ultrasound is messy. Speckle, shadows, attenuation.
Most pretraining objectives end up modeling the scanner, not the heart.
The idea:
Stop reconstructing pixels.
Predict latent structure instead.
EchoJEPA discards what’s unpredictable and locks onto what matters clinically:
➡️ chamber geometry
➡️ wall motion
➡️ valve dynamics
The results (frozen encoder, no fine-tuning):
• 20% ↓ error in LVEF
• 17% ↓ error in RVSP
• 79% accuracy with 1% labels (vs 42% for baselines w/ 100%)
• 2% degradation under acoustic artifacts (vs 17%)
• Zero-shot pediatric transfer beats all fine-tuned models
Why this works:
When we project embeddings:
❌ prior methods → diffuse, entangled clusters
✅ EchoJEPA → clean anatomical organization
Structure separated from acquisition noise.
📄 Paper: arxiv.org/abs/2602.02603
💻 Code: github.com/bowang-lab/EchoJE…
Huge credit to @alifmunim , who pushed JEPA thinking into medical video and led this effort 💥
Guidance from @AIatMeta (@garridoq_, @koustuvsinha)
Co-authors: @adibvafa @TeodoraSzasz @A_Attarpour @riverjiang @JSlivnickMD
Teams: @UHN @awscloud @UofT @UChicago @UCSF @PhilipsHealth
This is representation learning for physiology, not pixels.
Maciej Tysarowski retweeted
“Claude is down.
time to go back to hand coding.”
This quoted post is unavailable.
Maciej Tysarowski retweeted
If you are not blown away by the power of OpenEvidence AI then you are not using it correctly.
I recently tried some pretty complex cases and it provided a perfect differential, next diagnostic steps, areas of uncertainty, and follow up plan.
Maciej Tysarowski retweeted
A medical malpractice case about communication failures around imaging: 🧵
A young, very tall man with chest pain, nausea, and abdominal pain presented to the ED and was worked up by the emergency physician.
Maciej Tysarowski retweeted
Recruiting Advanced #yesCCT #whyCMR imaging fellows for 2027-28. Division of Cardiovascular Imaging @StFrancis_LI @purviparwani @onco_cardiology @DrRyanPDaly @iamritu
Maciej Tysarowski retweeted
Insane stuff going down in Lisbon right now
I’m at the Europe’s biggest startup conference
Germany’s latest unicorn is presenting on stage
€112,000 annual revenue in just 6 years
Suddenly, someone in the audience stands up and asks a loud question
“Are you planning to expand to the US?”
The crowd gasped
CEO laughed and said “I’m not interested in third world markets”
The room exploded with ovation
Someone played a European Union anthem on their Siemens cell phone
After the panel I saw three German VCs fighting to offer this company €50k at a €125k valuation
I’ve never been more bullish on Europe
Maciej Tysarowski retweeted
Another NotebookLM upgrade - in under the wire - Gemini 3.0!
I've been testing it in NLM chat and am finding answers to be:
1. significantly more nuanced
2. long context is way, way more accurate than the NotebookLM of a month ago
3. a more pleasant vibe? send feedback on this one!
NotebookLM at the frontier of model performance and long context!
Maciej Tysarowski retweeted
For any cardiology fellows interested in training in multimodality CV imaging --> we have an unexpected opening for our 2 year fellowship starting July 1, 2026
Please RT/share #YesCCT #ThinkPET #WhyCMR #EchoFirst #CVImaging #ACCImaging #PCCT
Maciej Tysarowski retweeted
Look forward to sharing our science led by @Cross_JL & @shawnwahi @YaleMed @YaleCardiology @YaleIMed @YaleCancer @escardio 2025 in Madrid Spain! #AI #LLM #CMR & #ML models to predict ICI cardiotoxicity w/new risk calculator @attilafehermd @ibolya_csecs @mtysar @MariamBaigMD
Maciej Tysarowski retweeted
We're accepting applications for our 2026-2027 Cardio-Oncology Fellowship, which includes a focus in advanced cardiac imaging training. Learn more and apply: medicine.yale.edu/internal-m…
@CardioOnc_MR_CT #cardiooncology #cardiotwitter


































