LAB ACTIVITY

News

Recent publications, awards, press, and milestones from the lab.

2026-10-13

Adult AI models adapted for newborns in Nature Biomedical Engineering

New Nature Biomedical Engineering study led by Vishwanatha Rao, Emma Chen, and Agustina Saenz with 30 neonatal intensive care units in 20 countries and territories through MAIDA, Medical AI Data for All. Starting from a model trained on adult ICU X-rays, each NICU retrained it on 50 of its own newborn X-rays and passed it to the next hospital, without sharing patient images. Breathing-tube placement error fell at 29 of 30 hospitals. Code and model weights are open.

2026-09-08

Three papers in the Machine Learning for Healthcare 2026 proceedings

Xucheng Wang and colleagues show that a low-cost robot arm, trained on under an hour of demonstrations, followed 23 of 25 stitches alongside a surgeon in open surgery. Oishi Banerjee and Alexandra Willauer test whether generated patient images are realistic and clinically accurate enough for a virtual physical exam. Emma Chen and colleagues find that asking frontier LLMs for a full survival curve ranks and calibrates patient risk better than asking for a single date.

2026-05-01

ReXGrounding Challenge at MICCAI 2026

Launching the ReXGrounding Challenge — grounding free-text radiology findings to 3D CT segmentations on 3,492 expert-annotated chest CT scans across 14 focal and non-focal abnormality categories. Pre-registration is open; results presented at MICCAI 2026.

2026-04-08

Head CT features improve cardiovascular risk prediction in JACC

New JACC study with co-first author Julián Acosta: deep learning features from routine head CTs improve cardiovascular risk prediction over clinical-only models (C-index 0.82 vs. 0.75) and reclassify 15.7% of ED patients — most correctly — by surfacing subclinical vascular disease.

2026-03-20

Generalist biological AI in Nature Biotechnology

New Nature Biotechnology perspective led by Vish Rao on generalist biological AI — arguing the next frontier is not better single-task models, but systems that operate across the many layers of biology.

2026-03-12

Clinical Environment Simulator in Nature Medicine

New Nature Medicine perspective (Luo et al.) proposing the Clinical Environment Simulator: evaluate medical AI inside a digital hospital where every decision dynamically alters future patient states, instead of relying on static benchmarks.

2026-03-05

MedVersa generalist medical foundation model in NEJM AI

New in NEJM AI: MedVersa, an LLM-orchestrated generalist foundation model that outperforms specialist systems across classification, detection, and reporting — and that radiologists judged clinically equivalent to human-written chest X-ray reports 64% of the time (91% for normal studies).