Please login or create an account. If you do not have access to this content, you will be shown a 30 second preview and licensing options.
- Presentation
Technology and AI Tools in Dermatologic Surgery
Description
The speaker, a Mohs surgeon, reviews practical technologies and AI tools relevant to dermatologic surgery. He highlights OpenEvidence as a vetted, HIPAA-compliant LLM that draws from curated medical sources such as trials, reviews, guidelines, and drug labels, making it useful for literature review, differentials, image-based suggestions, and prior authorization letters, while warning that it may lag behind the newest publications and that other dermatology LLMs require caution and verification. He then discusses ambient AI documentation tools, which listen to clinic encounters and generate notes; evidence suggests mixed effects on documentation time but meaningful improvements in burnout, task load, and patient engagement, especially in complex visits, though accuracy, editing burden, language limitations, EHR integration, and cost remain concerns. He also presents practical apps for cutaneous squamous cell carcinoma, including a personalized risk calculator based on large multi-center data and a preview of COMPASS, a proposed new staging system that aims to integrate staging tools and resources. Switching to pathology, he reviews AI for frozen sections in basal cell carcinoma: whole-slide classification performs well, but segmentation remains inconsistent and is not yet clinically ready, with performance varying by subtype and false positives/negatives still common. Finally, he offers a low-tech communication tip: headset systems, which are inexpensive and improve team communication and workflow, though they require some learning and can introduce noise or miscommunication.
View moreConclusions
- OpenEvidence appears to be a safer and more clinically useful LLM than general-purpose chatbots because it relies on vetted medical sources, includes guardrails, and is HIPAA compliant when used appropriately.
- Ambient AI documentation can reduce clinician burnout, task load, and sometimes note-writing time, but its effect on documentation efficiency is mixed and varies by specialty.
- AI-generated clinical notes are generally judged to be comparable in quality to manually written notes, but they still require careful human editing and do not yet work well for every note type.
- The greatest practical benefit of ambient scribes may be in complex, conversation-heavy encounters where they can draft most of the note and reduce after-hours charting burden.
- The usefulness of ambient AI is not uniform across specialties, with primary care benefiting more consistently than medical or surgical subspecialties.
- Personalized risk calculators for cutaneous squamous cell carcinoma can help guide treatment decisions most effectively in intermediate-risk cases where staging alone is less informative.
- The proposed COMPASS staging system aims to unify and improve prognostication for cutaneous squamous cell carcinoma by integrating multiple staging frameworks and tumor features.
- AI for frozen-section pathology currently performs well for whole-slide classification of basal cell carcinoma, but that task is less clinically useful than precise tumor localization.
- Segmentation of basal cell carcinoma on Mohs frozen sections remains imperfect, with false positives and false negatives limiting current clinical deployment.
- Certain basal cell carcinoma subtypes, especially micronodular and infiltrative patterns, are harder for AI models to detect than nodular or superficial subtypes.
- A more clinically meaningful future direction for pathology AI is outcome-based assistance that aligns model output with Mohs maps and surgeon decision-making rather than abstract segmentation metrics.
- Simple communication headsets remain a practical, inexpensive way to improve teamwork and workflow in dermatologic surgery despite some usability drawbacks.
- Varra et al: Sn/Sp .71/.75
- Van Zon et al: DICE score 0.66
- Bonnefille et al: 96% of BCC foci (843/877)
- Tan et al: AUC .943, DICE 0.90
- Ambient AI Scribes in Clinical Practice: A Randomized Trial#10.1056/aioa2501000
- Evaluation of an Ambient Artificial Intelligence Documentation Platform for Clinicians#10.1001/jamanetworkopen.2025.8614