@cardslab.bsky.social is leading a workshop at #AHA25 #QCOR25

#AI in Action: LLM-Powered Quality Assessment & Improvement in the EHR

Learn how to:
🔍 Use LLMs to measure care quality
⚡ Build real-time AI workflows
🌐 Implement scalable, equitable solutions

📍 Session: QCOR.01 | 💻 Bring your laptop!

🚨 The @cardslab.bsky.social at Yale is recruiting postdocs in our clinician–data scientist track!

🔹 MD or MD/PhD
🔹 Python/R skills
🔹 Research experience
🔹 Passion for AI + digital health

📝 Apply early: CV + 250-word statement
🔗 apply.cards-lab.org/apply/CATSP004

#AI #Cardiology #Postdoc

CarDS Lab (@cardslab.bsky.social) will be in full force at #ACC25. We are presenting our work spanning:

#AI for cardiac diagnosis
#DigitalBiomarkers
#Precision inference from RCTs
#DigitalHealthPolicy
#NLP & #LLM #RAG for scalable phenotyping in the #EHR

See you in Chicago this weekend!

Also, read the excellent editorial that accompanies our study.

The single image-based AI-ECG score was either comparable or exceeded the performance of a multicomponent standard in PCP-HF (computable at Yale and ELSA-Brasil)

We demonstrate a several-fold higher risk of HF with higher thresholds of AI-ECG risk score for LV dysfunction.

The effect estimates were more pronounced in rigorous cohort studies in the UK and Brazil, likely due to more complete & robust outcome capture vs EHR

Study design⬇️. We posited that cross-sectional signatures of LV dysfunction on AI-ECG would portend future HF risk.

Rather than build models for HF risk with time-varying confounding, we focused on a mechanistic approach.

+ broadly validated the result vs. risk scores

The work is a collaboration between our growing global Precardia network - a network of federated science in AI. Watch this space for more: precardia.org

Tagging those on here: @ekoikonomou.bsky.social @aline-pedroso.bsky.social

In the European Heart Journal, we report a novel paradigm for predicting HF risk.

AI applied to a single ECG image. Without risk scores or other testing.

Led by @lovedeepdhingra.bsky.social @cardslab.bsky.social

@yaleschoolofmed.bsky.social

🔗 academic.oup.com/eurheartj/ad...

Happy New Year from my friends & work family at @cardslab.bsky.social

Grateful for 2024, where we grew our group & strengthened our culture & community around advancing science while having fun 😊

Hoping to build on this foundation in 2025!

If you’re at #CVCT24, join us for:
1️⃣AI for design and dissemination of clinical trials (Mon, 12/9) w @hmkyale.bsky.social @jonwcunningham.bsky.social

2️⃣Enabling screening and tracking of cardiac amyloidosis (Tue, 12/10)

@cardslab.bsky.social

Schedule:
www.globalcvctforum.com/_files/ugd/d...

The model generalized to external datasets EchoNet-LVH and EchoNet-Dynamic to achieve excellent performance in these new datasets it had never seen.
Also, it serves as a foundation model for new tasks, with the ability to efficiently fine-tune

The model learned task-specific views - demonstrating that it correctly identifies the relevant views for each task and weighs them as an expert human reader would.

The model achieved excellent performance on the entire range of tasks for both full TTEs but also abbreviated protocols obtainable on handheld devices

The model uses a supervised learning strategy, combining an image encoder to examine frames of echo videos, and capturing temporal information across video frames using a transformer model & then performing each of the tasks automatically

We used 1.1 M Echo videos from nearly 30K studies to build a model that does inference on 39 tasks - from measurement to quantification to classification.

The objective: to build a tool that can automatically leverage any echo study & make complete inference without human input

Presented at #AHA24 Late Breaking session, we are pleased to announce #PanEcho - Complete AI-enabled echocardiography interpretation with multi-task deep learning.

Led by @giholste.bsky.social & @ekoikonomou.bsky.social
@cardslab.bsky.social

Preprint, code, and model: cards-lab.org/panecho