Cornell AI model spots advanced heart failure

This is the kind of medical AI result that actually matters: not a flashy demo, but a workflow shortcut that could widen access to a scarce, high-friction test.

• –The model turns routine echo plus EHR data into a proxy for CPET, which is a much more deployable path than sending patients to specialized exercise labs.
• –The paper reports stronger performance than prior work and an external validation boost, which is a good sign this is more than an in-sample curiosity.
• –The biggest caveat is still generalization: the training and validation data come from one health-system ecosystem, so multi-site prospective testing is the real gatekeeper.
• –The open GitHub repo with code and weights is a plus for reproducibility, which is still too rare in clinical ML.
• –If this holds up prospectively, the real win is earlier referral to advanced heart failure care, not just better model metrics.