Surgical Innovations

Paramedics call ahead during trauma transport. That call is often the only information the hospital gets. More than 98% of the words in these calls are nonmedical. The rest gets lost in noise, static, and time pressure. A new study tested an AI model on pediatric trauma calls. It stripped the noise and compressed each transcript by 80%. Accuracy matched trauma staff. Here's the real finding. When trauma staff were exposed to the AI's read after making an incorrect triage deci

University of Bern A wound gets sutured. The surgery is over. But the riskiest hours haven't started yet. In Bern, researchers developed an AI model named CARESCORE that analyzes vital signs collected during surgery, including blood pressure, heart rate, and oxygen saturation. Within seconds of the last stitch, it calculates a patient's infection risk. Trained on over 10,000 surgeries. More accurate than models that only look at pre-op data. It won't replace a surgeon's judgm

Source: NVIDIA For years, AI in surgery lived on screens. It read scans. It predicted outcomes. Now it steps into the room. NVIDIA just released open physical AI models built specifically for healthcare robotics. Open H is the largest surgical robotics dataset ever assembled, with over 700 hours of real surgical video. Cosmos H generates synthetic surgical scenes to train robots faster. GR00T H turns spoken clinical commands into robot motion. CMR Surgical, Johnson & Johnson

Illustration by Alicia Tatone. Source: Getty. 80% of physicians are already using AI at work. Most of the generative AI tools they use have not been approved by the FDA. A Science study found ChatGPT more accurate at diagnosis than hundreds of practicing doctors. Even the Harvard physician who led the research expressed concern about what the results might mean. Medicine likes to think it moves at the speed of evidence. In reality, it often moves at the speed of utility. AI i

Source: Noah Berger Women with abnormal mammograms often wait weeks to find out if they have cancer. UCSF used an AI model called Mirai, trained on hundreds of thousands of mammograms, to identify the 12.7% of patients most likely to have cancer. Those patients went from screening to evaluation to biopsy in a single day. Wait time for a diagnostic evaluation dropped from several weeks to about an hour. The wait time for a biopsy dropped from two months to under 10 days. The A

Source: Stanford Medicine A discharge summary takes 15 to 30 minutes to write. Stanford built an AI agent that generates one automatically every morning for each patient. Physicians used it. Burnout scores dropped. Hallucinations were rare at 2%. The tool was not designed to replace clinical judgment. It was designed to eliminate the part of the job that exhausts physicians before they even begin. Read more: Stanford Medicine