Phenotyping of AMD Patients Using Artificial Intelligence-Driven Biomarker Patterns in Optical Coherence Tomography

AbstractPurpose To leverage artificial intelligence (AI)-based optical coherence tomography (OCT) analysis to classify age-related macular degeneration (AMD) images into distinct subgroups based on retinal layer and fluid biomarkers. Design Retrospective analysis of a dataset of retinal OCT images. Participants Anonymized data from 157 AMD patients and 58 healthy volunteers. Methods This study analyzed OCT scans from AMD patients and healthy volunteers to extract retinal layer thickness and fluid biomarkers by AI-driven image segmentation. We performed dimensionality reduction using parametric Uniform Manifold Approximation and Projection (UMAP), followed by k-means clustering to partition the continuous disease space into nine subphenotypes for analysis. Statistical analyses included linear mixed-effects modeling, analysis of variance (ANOVA), and Fisher's exact tests to assess biomarker differences, visual acuity (VA) and lesion size across clusters. We mapped International Classification of Diseases, 10th Revision (ICD-10) codes to validate disease staging. Main Outcome Measures We examined the potential of AI-driven biomarker patterns based on the comparison to traditional clinical measures such as visual acuity, lesion size and ICD-10 codes for different disease stages. Results By dividing the continuous landscape into nine regions for interpretability, we observed that each cluster was characterized by a distinct biomarker pattern. Clusters exhibited outer retinal thinning of varying degrees and showed significant differences in inner retinal layers, choroidal thickness, sub-retinal pigment epithelium material and fluid presence compared to a healthy cohort, suggesting diverse AMD subtypes beyond simple characterization by VA and atrophic lesion size alone. The distribution of ICD-10 codes further supported the interpretation of the AMD landscape as a representation of AMD progression. Conclusions This study reveals a continuous AMD landscape based on OCT-derived biomarkers, suggesting the existence of subphenotypes beyond traditional staging methods. The findings further highlight the limitations of fundus-based disease classification and support the role of OCT scans in personalized disease management.