Ex vivo fluorescence confocal microscopy for real-time margin assessment of nonmelanoma skin cancer in Mohs surgery: A prospective interventional study

New-generation ex vivo fluorescence confocal microscopy (EVFCM) generates hematoxylin-eosin-like images directly comparable to conventional histopathology. Its interobserver performance with a novel acquisition mode in non-melanoma skin cancer (NMSC) micrographically controlled surgery remains to be defined. In this prospective interventional study, 42 tumors from 39 patients undergoing Mohs surgery between January and October 2025 were examined with EVFCM. Specimens were orthogonally sectioned, stained, and acquired in digital-stain mode. Peripheral and deep margins were independently and blindly reviewed by a dermatologist and a pathologist, both trained in EVFCM interpretation. Concordant negative margins were considered tumor-free, whereas positive, discordant, or equivocal findings prompted re-excision and re-evaluation. Diagnostic agreement with histopathology was assessed using Cohen's κ statistic. A total of 261 margins were evaluated. Inter-reader agreement between the dermatologist and pathologist was almost perfect for both margin assessment (κ=0.863; SE 0.067; p<0.001; observed agreement 98.6%) and histotype assessment (κ=0.882; SE 0.031; p<0.001; observed agreement 97.6%). Compared with histopathology, EVFCM showed high performance, with an overall diagnostic accuracy of 96.4%. The AUC for the combined assessment was 0.875 (SE 0.074; 95% CI 0.730-1.000). Only two discrepancies were observed between EVFCM and histopathology: one false-negative margin assessment and one discordance attributable to differences in sample processing. These findings support EVFCM as a reliable tool for intraoperative margin assessment in NMSC. Comparable performance between dermatologist and pathologist readers supports its potential integration into Mohs surgery workflows; however, potential workflow advantages of the standardized sample-processing approach require prospective validation using quantitative time metrics.