AI-Assisted Basal Cell Carcinoma Analysis in Line-Field Confocal OCT: Exploring Diagnostic Metrics

Artificial Intelligence (AI)-assisted line-field confocal optical coherence tomography (LC-OCT) may support the clinical diagnosis of basal cell carcinoma (BCC). The current model produces frame-level BCC scores (0%-100%), but interpretation is challenging due to false positive frames. We present two novel scan-level metrics: the composite score (percentage of frames scoring 100% per scan) and the stability score (longest sequence of consecutive 100%-score-frames per scan). LC-OCT videos (n = 568) from 82 lesions from 52 patients were analyzed, including 18 BCCs and 64 non-BCCs. Both scores were significantly higher for BCCs than for non-BCCs (p < 0.0001). An exploratory threshold of 48 consecutive frames with 100% AI scores yielded 100% sensitivity (95% CI: 81.5%-100%) and 100% specificity (95% CI: 94.4%-100%). In conclusion, scan-level composite and stability scores demonstrate consistent separation between BCC and non-BCC lesions; however, external validation in independent cohorts is required before clinical implementation.