Universal Wavelet Units in 3D Retinal Layer Segmentation

2025-07-21

AI Summary

Traditional max-pooling in 3D OCT volume segmentation degrades spatial detail and blurs inter-layer boundaries. To address this, we propose embedding tunable wavelet units (UwUs) into a 3D retinal layer segmentation network, introducing— for the first time—learnable wavelet filter banks to replace fixed downsampling operations. We design three wavelet-based downsampling modules: OrthLattUwU, BiorthLattUwU, and LS-BiorthLattUwU, and integrate them into a motion-corrected MGU-Net architecture to enable joint modeling and multi-scale fusion of high-frequency textural and low-frequency semantic features. Evaluated on the Jacobs Retina Center dataset, our approach significantly improves segmentation accuracy and Dice scores; LS-BiorthLattUwU achieves the best performance, demonstrating that learnable wavelet downsampling provides critical gains in structural fidelity and spatial consistency for medical volumetric image segmentation.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Multimodal Learning
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

This paper presents the first study to apply tunable wavelet units (UwUs) for 3D retinal layer segmentation from Optical Coherence Tomography (OCT) volumes. To overcome the limitations of conventional max-pooling, we integrate three wavelet-based downsampling modules, OrthLattUwU, BiorthLattUwU, and LS-BiorthLattUwU, into a motion-corrected MGU-Net architecture. These modules use learnable lattice filter banks to preserve both low- and high-frequency features, enhancing spatial detail and structural consistency. Evaluated on the Jacobs Retina Center (JRC) OCT dataset, our framework shows significant improvement in accuracy and Dice score, particularly with LS-BiorthLattUwU, highlighting the benefits of tunable wavelet filters in volumetric medical image segmentation.

Problem

• Applying tunable wavelet units for 3D retinal layer segmentation
• Overcoming limitations of conventional max-pooling in OCT volumes
• Enhancing spatial detail and structural consistency in segmentation

Innovation

• Tunable wavelet units for 3D retinal segmentation
• Wavelet-based downsampling modules in MGU-Net
• Learnable lattice filter banks preserve frequency features