AI Augmented Confocal Laser Endomicroscopy for Rapid Intraoperative Diagnosis of Brain Tumors

초록

Intraoperative diagnosis is important for brain tumor surgery. Although frozen section (FS) remains the standard, its time demands and logistical limitations hinder real-time decision-making. Confocal laser endomicroscopy (CLE) offers rapid high-resolution imaging for efficient diagnosis. This multicenter prospective trial evaluated the diagnostic performance of CLE while developing a novel Swin Transformer–based AI diagnostic model. A total of 461 biopsies from 376 patients were analyzed. CLE demonstrated non-inferior diagnostic accuracy to FS (0.94 vs 0.92; P = 0.14), with comparable sensitivity (0.96 vs 0.95; P = 0.40) and specificity (0.79 vs 0.68; P = 0.31). CLE achieved a significantly faster median turnaround (5 m 56 s vs 20 m; P < 0.001). The AI model showed diagnostic accuracies of 0.94 for tumor detection and 0.88 for biopsy subtype diagnosis. Overall, CLE demonstrated diagnostic efficacy comparable to FS with significantly shorter turnaround. The AI diagnostic model also showed promising results supporting its potential integration with the CLE platform.