OCT-derived SubRPE Compartment Radiomics Features are Associated with the Development of Subfoveal Geographic Atrophy

Abstract

Purpose : Identifying patients who are at high risk for development of subfoveal geographic atrophy (sfGA) in dry age-related macular degeneration (AMD) will be critical to personalized treatment management and potential clinical trial enrichment for eyes at greatest risk. In this study, the role of shape and texture-based radiomic features within the sub-retinal pigment epithelium (subRPE, i.e., space bounded by RPE and Bruch’s membrane) on baseline SD-OCT scans is evaluated as potential biomarkers to discriminate between eyes that are at highest risk of sfGA progression.

Methods : This was an IRB-approved retrospective image analysis study of subjects with dry AMD without sfGA with 5-year clinical and SD-OCT follow-up. Based on sfGA status at year five, eyes were categorized as Progressors and Non-progressors. A total of 26 shape-based fractal dimension (FD) and 364 texture-based radiomics features were extracted from the subRPE compartment of the baseline SD-OCT scans. Minimum Redundancy Maximum Relevance (MRMR) feature selection method was employed to identify features from the training set (N=96) and evaluated with 3 different machine-learning classifiers over 500 iterations of 3-fold cross validation. Classifier performance was validated on the test set (N=41).

Results : Random Forest classifier yielded a cross-validated area under the Receiver Operating Characteristics curve (AUC) of 0.89±0.09 and 0.76±0.06 using isolated shape-based FD and isolated texture-based radiomics features, respectively. Mean fractal entropy was identified as the most significant shape-based biomarker with higher values of entropy being associated with greater shape disorder and risk for sfGA progression. The Laws texture features were identified as the most discriminating features with the Progressors having higher feature expression. This suggests higher degree of heterogeneity within the subRPE compartment texture for the Progressors. A combination of shape and texture features yielded a significant improvement in the classifier performance with AUC=0.92±0.02.

Conclusions : In this study, radiomics-based characterization of the subRPE compartment in dry AMD identified multiple potential imaging biomarkers. Future works will involve prospective multi-institutional validation of the fractal and texture features for clinical trial enrichment and assessments for therapeutic response in dry AMD patients.

This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.