Identifying MRI-Defined Optic Pathway Glioma in NF1 Using Macular OCT and Machine Learning.
Authors
Affiliations (3)
Affiliations (3)
- Computer Science Department, Jerusalem Multidisciplinary College, Jerusalem, Israel.
- Fundación Oftalmológica del Mediterráneo, Valencia, Spain.
- Department of Optics and Optometry, Vision, Optometry and Health Research Group (VOS), Universitat Politècnica de Catalunya-BarcelonaTech (UPC), Barcelona, Spain.
Abstract
To evaluate whether macular optical coherence tomography (OCT) distinguishes eyes with versus without recorded magnetic resonance imaging (MRI)-defined eye-level optic pathway glioma (OPG) status in neurofibromatosis type 1, and whether longitudinal OCT, visual/clinical variables, optic-nerve, or single-feature OCT add value. One study eye was selected per participant independently of OPG status. The retrospective cohort included 182 participants (58 OPG-positive and 124 OPG-negative eyes). Six primary feature sets compared baseline and longitudinal macular OCT, with and without clinical/visual variables, plus clinical/visual variables alone and visual acuity (VA) alone. Paired macular versus optic-nerve analyses used 141 participants with at least two paired examination dates. Baseline macular OCT achieved pooled out-of-fold area under the precision-recall curve (PR-AUC) 0.819 and area under the receiver operating characteristic curve (ROC-AUC) 0.870, versus 0.551 and 0.667 for VA alone. Longitudinal summaries and added clinical/visual variables provided no clear gain. Baseline macular and optic-nerve OCT showed similar discrimination in the paired cohort (PR-AUC 0.787 vs. 0.802). Macular ganglion cell complex (GCL++) alone achieved PR-AUC 0.819 and ROC-AUC 0.844. Of 58 OPG-positive eyes, 31 (53.4%) had preserved baseline VA; among participants with preserved VA, baseline macular OCT retained discrimination (PR-AUC 0.701; ROC-AUC 0.831). Macular OCT provided structural information beyond VA and showed similar discrimination to optic-nerve OCT. Longitudinal summaries offered no clear gain over baseline OCT, while GCL++ alone showed strong discrimination. Simple macular OCT measures may complement clinical assessment and MRI surveillance but require external validation before clinical use.