Retinal layer segmentation in Optical Coherence Tomography (OCT) is crucial for extracting quantitative biomarkers of retinal structure, especially in the analysis of neurodegenerative diseases.
However, challenges arise from speckle noise, shadowing artifacts, low contrast between adjacent layers, anatomical variability across subjects, and domain shifts due to different acquisition protocols and clinical populations. While deep learning methods have shown remarkable performance, their robustness and generalization across heterogeneous datasets remain limited.
This work investigates the role of spatial normalization as a preprocessing strategy to mitigate geometric domain shifts and enhance the consistency of retinal layer segmentation. We introduce a fovea-centered normalization framework that aligns OCT volumes into a common anatomical reference.
A comprehensive evaluation of state-of-the-art deep learning architectures is performed, combining conventional overlap-based metrics at the B-scan level with topology-aware metrics at the A-scan level and thickness-based measures at the en-face level.
In scenarios where ground truth is unavailable, we propose topology violation quantitative metrics that do not require ground truth annotations, alongside a thickness-based qualitative assessment that captures structural consistency and clinically relevant patterns at the en-face level.
The results highlight the significance of spatial normalization in OCT segmentation pipelines, facilitating the development of robust and clinically meaningful retinal analysis tools, enabling reliable biomarker extraction and downstream computational analysis in neurodegenerative research.
Blogger's Review: This paper delves into the application of spatial normalization in OCT retinal layer segmentation, offering an effective solution to traditional methods' limitations. Particularly in the context of neurodegenerative diseases, it enhances segmentation consistency and reliability, providing crucial theoretical support and practical foundations for future clinical applications.