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

Area of Science:

• Dermatology
• Biomedical Engineering
• Computational Pathology

Background:

• Accurate assessment of surgical margins is critical in Mohs micrographic surgery for basal cell carcinoma (BCC) removal.
• Traditional histopathology can be time-consuming and may involve tissue loss.
• Developing adjunct tools to improve margin assessment is an ongoing area of research.

Purpose of the Study:

• To evaluate a label-free, high-frequency bioelectrical impedance spectroscopy (HF-BIS) method combined with machine learning (ML) as an adjunct to histopathology.
• To generate probability heat maps for estimating basal cell carcinoma risk at surgical margins.
• To assess the feasibility of integrating this technology into the Mohs surgery workflow.

Main Methods:

• A two-step cascade machine-learning algorithm was applied to data from 98 dermal specimens of nodular BCC from 55 patients.
• High-frequency bioelectrical impedance spectroscopy was used to generate data from excised dermal specimens.
• Spatial-tolerance scoring was implemented to account for potential label noise from manual histology misregistration.

Main Results:

• The ML algorithm achieved an ROC AUC of 0.878 ± 0.048 in identifying cancer-positive tissue locations.
• With spatial-tolerance scoring, the performance improved to an ROC AUC of 0.993 ± 0.003.
• High sensitivity (95.6% ± 4.7%) and specificity (96.1% ± 5.0%) were achieved, demonstrating excellent diagnostic accuracy.

Conclusions:

• Label-free HF-BIS coupled with ML is a promising non-destructive tool for real-time margin assessment in Mohs surgery.
• The technology demonstrated high accuracy in identifying basal cell carcinoma at surgical margins.
• This method offers rapid, integrated surgical guidance, potentially preserving frozen sections for definitive assessment.