
← Behind The Knife: The Surgery Podcast9 nov · 35 min
Clinical Informatics for Surgeons, Episode 3: AI in the OR and Beyond; Machines That See, Read, and Act
Picture three machines in your next operation. One watches the laparoscopic feed and flags the zone that is not safe to divide. One reads the chart and has the op note and a plain-language patient summary done before you close. One works quietly in the background and has the ICU bed request queued before anyone asks. Notice that none of them hands you a percentage. In the series finale, Drs. Mason Pearce, Tyler Loftus, and Dan Hashimoto trace surgical AI from the NSQIP risk calculator to computer vision, large language models, and agentic systems: which of these is already in operating rooms, which is arriving fast, which is still mostly promise, and how a surgeon without a machine learning degree stays in control of all three.
Hosts:
Mason Pearce, MD, General Surgery Resident and Clinical Informatics Fellow, University of South Florida; Behind the Knife Education Fellow
Tyler J. Loftus, MD, PhD, Acute Care Surgeon and Associate Professor of Surgery; Program Director, General Surgery Residency, University of Florida; board certified in Clinical Informatics; Past Chair, ACS Health IT Committee
Daniel A. Hashimoto, MD, MSTR, Associate Professor of Surgery and Director of Surgical Informatics, Duke University Health System; editor of Artificial Intelligence in Surgery
Learning objectives:
Distinguish predictive, generative, and agentic AI (models that score, read and write, or act) and give a surgical example of each.
Trace surgical risk prediction from the regression-based ACS NSQIP calculator to machine learning models (XGBoost, POTTER, MySurgeryRisk), and explain why external, multicenter validation matters.
Describe how computer vision recognizes operative steps and safe versus unsafe dissection zones, and why standardized annotation and real-world video are needed for those models to hold up in practice.
Identify the strengths and failure modes of large language models (fluent but wrong output, automation bias) in documentation, perioperative triage, and value-informed proxy decision support.
Explain what makes an AI system agentic, how errors can compound across steps, and where human oversight belongs.
Apply a practical checklist to any surgical AI tool (data source, validation on patients like yours, drift, explainability, effect on decisions and outcomes) and seek a seat on hospital AI and clinical decision support committees.