195: Ultrasound AI Outperforms Surgeons Diagnosing Burns

Send us Fan MailPaper Discussed in this AI Journal Club:Masry ME, Gnyawali S, Jacobson M, Xue Y, Sen C, Wachs J, Gordillo G. AutoMated Burn Diagnostic System for Healthcare (AMBUSH). Plast Reconstr Surg Glob Open. 2023 Oct 18;11(10 Suppl):128-129. doi: 10.1097/01.GOX.0000992564.42240.e3. PMCID: PMC10566867.Episode Summary: In this journal club deep dive, we tackle a clinical problem that has frustrated surgeons for decades: accurately diagnosing burn depth. We examine a groundbreaking study introducing AMBUSH-AI, an artificial intelligence system that evaluates ultrasound imaging to outperform the diagnostic accuracy of human experts. When the stakes are a lifetime of severe scarring from under-treatment versus the painful trauma of an unnecessary skin graft, can a combination of standard ultrasound and AI completely eliminate the dangerous guesswork of human visual inspection?.In This Episode, We Cover:• The Diagnostic "Coin Toss": Why distinguishing between deep partial and full-thickness (third-degree) burns is the ultimate clinical challenge. We discuss the terrifying reality that experienced burn surgeons only achieve about 76% accuracy in visual assessments, while non-experts sit at 50%—literally a coin toss.• The Two-Part Tech Combo (Anatomy and Stiffness): ◦ B-Mode Ultrasound: The standard imaging modality that provides the anatomical landmarks, letting the machine know exactly where the epidermis, dermis, and hypodermis are located. ◦ Tissue Doppler Elastography Imaging (TDI): The secret sauce that measures tissue stiffness. When skin burns, structural proteins denature and tangle, making the tissue physically stiffer. TDI visualizes this stiffness with color overlays—red for healthy and supple, blue for stiff and burned.• Finding the Ground Truth: Why you can't calibrate a new, precise tool against a broken ruler. Instead of comparing the AI to flawed human visual estimates, the researchers validated the AI against actual tissue biopsies taken in the operating room, establishing an undeniable histological reality.• The Results - Outperforming the Experts: In the human clinical trial, AMBUSH-AI achieved a staggering 95% overall accuracy. Crucially, it had a 100% sensitivity rate for surgical cases, meaning it did not miss a single patient who definitively needed an operation to prevent severe morbidity.• The AI's "Glass Box" Design: Why surgeons will never trust a mysterious "black box" that just spits out a diagnosis. AMBUSH-AI is designed as an explainable model; it outputs plain text explaining its reasoning (e.g., "dominant, continuous blue pattern is present in the hypodermis"), acting as a transparent decision-support tool rather than a robot replacement.• The Future of Triage: How this technology could be paired with Point of Care Ultrasound (POCUS) on portable tablets in military combat zones and rural ERs, giving any medic or general doctor the diagnostic confidence of a 20-year burn specialist.Key Takeaway: The era of subjective, visual wound assessment is ending. By successfully translating the physical stiffness of a burn into objective, AI-interpreted data—a process called "computational palpation"—we can dramatically improve triage accuracy, save vital hospital resources, and spare patients from both dangerous under-treatment and unnecessary surgical trauma.Support the showGet the "Digital Pathology 101" FREE E-book and join us!