Artificial Intelligence-Based Quantification of Indocyanine Green Fluorescence for Standardized Perfusion Assessment in Endoscopic Skull Base Surgery

Abstract Indocyanine green (ICG) fluorescence angiography is used to assess nasoseptal flap (NSF) perfusion in endonasal skull base surgery (ESBS), but there is wide variability in how the ICG signal is measured and interpreted. Existing approaches rely on binary fluorescence presence, which lacks granularity, or manual clinician measurement, which is time-consuming. Artificial intelligence (AI) computer vision models may enable standardized, real-time fluorescence quantification. This study evaluated the ability of an AI model to measure ICG fluorescence during ESBS. A prospective study was conducted in adults undergoing ESBS, where posterior septal artery (PSA) perfusion was assessed with ICG at baseline and after tumor resection. If an NSF was harvested, ICG fluorescence was measured at the flap pedicle and body. Fluorescence intensity (FI) was quantified using a publicly available AI computer vision model (ChatGPT-4o) and compared with manual clinician measurements. Demographic, clinical, and surgical characteristics were analyzed for association with FI. A total of 44 patients were included. AI analysis did not significantly differ from manual FI at baseline (111.9 ± 32.4 vs. 115.8 ± 28.2, p = 0.545) or postresection (96.9 ± 38.3 vs. 110.1 ± 36.7, p = 0.133). Change from baseline to postresection was also similar (Δ −12.5 vs. Δ −3.2, p = 0.224). However, for the NSF body FI (n = 20), a significant difference was observed (48.5 ± 9.9 vs. 108.1 ± 31.0, p < 0.0001). AI-based ICG analysis closely matched clinician measurements for PSA and NSF perfusion, supporting feasibility for automated intraoperative assessment. Discrepancies in NSF body measurements likely reflect methodological or surface variability. Larger-scale validation is warranted.