Diagnostic Performance of Artificial Intelligence Corrected OCT Measurements in Highly Myopic Eyes with Glaucoma

Objectives: This study aimed to evaluate the diagnostic performance of peripapillary retinal nerve fiber layer (RNFL) thickness measurements corrected by artificial intelligence (AI) compared to original uncorrected values for glaucoma detection in highly myopic patients. Methods: This cross-sectional diagnostic accuracy study included 57 eyes from highly myopic patients (31 with glaucoma, 26 without glaucoma). Peripapillary RNFL parameters were obtained using Spectralis optical coherence tomography (OCT). A deep learning algorithm (MGU-Net) was employed to automatically segment retinal layers and compensate for scan tilt in elongated eyes, producing AI-corrected measurements. RNFL thickness values were extracted for six sectoral parameters (ST, SN, N, IN, T, IT) and global. Diagnostic performance was assessed using area under the ROC curve (AUC) and compared between corrected and uncorrected values. Multivariable logistic regression models were also developed using stepwise selection. Results: AI-corrected values were significantly lower than original measurements in all sectors (p < 0.001), with mean differences ranging from 15 to 35 µm. In glaucomatous eyes, significant thinning was observed in the global (p = 0.049) and inferior nasal (IN) sector (p = 0.037) among corrected values. The highest AUCs were found in IN (0.69), IT (0.67), and global (0.66) for corrected values, and in IT (0.63), T (0.59), and global (0.63) for uncorrected data. A model combining ST, T, and IT AI-corrected values achieved an AUC of 0.79. Conclusions: AI-corrected RNFL thickness measurements improve consistency and enhance diagnostic performance in highly myopic glaucoma patients. Correction algorithms may reduce false positives and help reveal glaucomatous damage otherwise obscured by myopic anatomical changes.