Related Experiment Videos
Development and validation of an artificial intelligence model for basal cell carcinoma detection and section
Li Yang Loo1, Wen-Bin Luo2, Isaac Hong-Kei Lam2
1Department of Dermatology, Singapore General Hospital, Singapore.
Background:
Mohs micrographic surgery for basal cell carcinoma (BCC) requires rapid intraoperative histologic interpretation. This is labor-intensive and incomplete tissue sections compromise margin assessment. Artificial intelligence has shown promise in BCC detection but section completeness assessment remains underexplored.
Objective:
Develop and validate an artificial intelligence model for BCC detection and section completeness assessment on Mohs frozen-section histology.
Methods:
We conducted a retrospective single-center cohort study using histology slides from 130 patients who underwent Mohs surgery for BCC from 2017 to 2025. Whole slides were digitized and annotated. A convolutional neural network model was trained to detect BCC and classify section completeness. Model performance assessment with standard performance metrics and region-of-interest evaluation was performed.
Results:
The BCC detection model achieved accuracy 0.993, sensitivity 0.915, specificity 0.996, precision 0.889, F1-score 0.902 at the tile level, with comparable region-of-interest sensitivity (0.910). The section completeness model achieved accuracy 0.890, sensitivity 0.948, specificity 0.524, precision 0.927, and F1-score 0.937.
Limitations:
Single-center retrospective design, limited representation of rare BCC subtypes, and no direct comparison with human experts.
Conclusion:
Although the specificity for section completeness was low, further research and additional model training may enable development of a model with improved performance.