Automated Retinal Layer and Fluid Segmentation and Cross-sectional Analysis using Spectral Domain Optical Coherence Tomography Images for Diabetic Retinopathy

2025-03-03

AI Summary

Manual segmentation of retinal layers and pathological fluids (e.g., cystoid edema, hyperreflective foci) in spectral-domain optical coherence tomography (SD-OCT) images of diabetic retinopathy (DR) patients is time-consuming and exhibits poor inter-rater reproducibility. Method: We propose the first AI framework jointly modeling ten retinal layers, cystoid macular edema, and hyperreflective foci. Our approach employs a comparative multi-model architecture based on SwinUNETR and VM-UNet, integrated with transfer learning, multi-scale feature fusion, and a clinically informed loss function incorporating anatomical priors. Contribution/Results: We uncover stage-specific (NPDR vs. PDR) retinal layer thickness alterations and establish interpretable, clinically validated correlations between key layer thicknesses and visual acuity. Quantitatively, SwinUNETR achieves state-of-the-art segmentation accuracy (mean Dice score = 0.92); derived thickness maps meet clinical utility standards. Annotation time is reduced by >90%, enabling robust DR staging, treatment response monitoring, and personalized management.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Deep Neural Architectures and Foundation Models
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

This study presents an AI-driven pipeline for automated retinal segmentation and thickness analysis in diabetic retinopathy (DR) using SD-OCT imaging. A deep neural network was trained to segment ten retinal layers, intra-retinal fluid, and hyperreflective foci (HRF), with performance evaluated across multiple architectures. SwinUNETR achieved the highest segmentation accuracy, while VM-Unet excelled in specific layers. Analysis revealed distinct thickness variations between NPDR and PDR, with correlations between layer thickness and visual acuity. The proposed method enhances DR assessment by reducing manual annotation effort and providing clinically relevant thickness maps for disease monitoring and treatment planning.

Problem

• Automated retinal layer and fluid segmentation in diabetic retinopathy.
• Cross-sectional analysis of retinal thickness using SD-OCT imaging.
• AI-driven pipeline for disease monitoring and treatment planning.

Innovation

• AI-driven pipeline for retinal segmentation
• Deep neural network for layer and fluid analysis
• SwinUNETR and VM-Unet for high accuracy segmentation