PanEcho turns routine heart ultrasounds into an instant, full-fledged cardiology report: trained on 1 million+ multi-view videos, this single network nails 39 measurements and diagnoses at once, posting a median AUC ≈ 0.91 and ~4 % error on ejection-fractio —comparable to expert reads.  Because it learns across all standard views, the same model works with a quick five-clip bedside scan and fine-tunes to new hospitals or paediatric hearts in just a few epochs, bringing expert-level echo interpretation to portable devices and underserviced settings

A single-view ultrasound clip is all this study’s AI needs to flag life-threatening aortic-valve narrowing: trained on 17,000+ echocardiography videos, the 3-D CNN ensemble spotted severe aortic stenosis with about 95–98 % accuracy across multiple hospitals and time periods. Saliency heat-maps highlight the valve and nearby structures — just where a cardiologist would look — so the system could bring expert-level screening to handheld point-of-care scanners in clinics or resource-limited settings.

A-CONECT replaces the trained human interviewer used in the I-CONECT dementia-prevention trial with a voice-enabled LLM chatbot that gently steers older adults with mild cognitive impairment into memory-rich, cognitively demanding conversations. We also create “digital-twin” versions of real patients to iterate on the bot offline, and show that compared with a vanilla ChatGPT interviewer, our software doubles memory-recall turns, boosts patient speech by ~65 %, yet costs barely a quarter of a human facilitator’s fee.