Universal Wavelet Units in 3D Retinal Layer Segmentation

TL;DR

This study introduces tunable wavelet units integrated into a neural network for improved 3D retinal layer segmentation in OCT images, demonstrating enhanced accuracy and detail preservation over traditional methods.

Contribution

It is the first to apply learnable wavelet-based downsampling modules within a neural network for 3D retinal segmentation, improving feature preservation and segmentation accuracy.

Findings

Significant accuracy improvement over baseline models

Enhanced preservation of spatial details in segmentation

LS-BiorthLattUwU achieved the highest Dice score

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.