Machine learning-guided bioelectrical impedance mapping for rapid adjunctive margin assessment in Mohs surgery

A label-free high-frequency bioelectrical impedance spectroscopy method, coupled with supervised machine learning, was evaluated as an adjunct to histopathology by generating probability heat maps of excised dermal specimens to estimate the risk of basal cell carcinoma at the surgical margin during Mohs micrographic surgery. In an IRB-approved study of 98 specimens from 55 patients with nodular BCC, a two-step cascade machine-learning algorithm identified cancer-positive tissue locations (ROC AUC 0.878 ± 0.048). To account for tumor-boundary label noise from manual histology misregistration, spatial-tolerance scoring yielded an ROC AUC of 0.993 ± 0.003, with sensitivity of 95.6% ± 4.7% and specificity of 96.1% ± 5.0%. Clinical relevance was addressed by recognizing that, with a median acquisition time of 3.7 min (IQR 2.6-5.5) per specimen, this technology is a non-destructive mapping tool that is easily integrated into the workflow to provide early surgical guidance while preserving frozen sections as the definitive clearance assessment.