Enhancing Cross-Modal Medical Image Segmentation through Compositionality

TL;DR

This paper introduces a compositionality-based approach for cross-modal medical image segmentation, improving interpretability, performance, and reducing complexity through learnable kernels that disentangle content and style in representations.

Contribution

It proposes a novel end-to-end framework using learnable von Mises-Fisher kernels to enforce compositionality, content-style disentanglement, and interpretability in cross-modal segmentation.

Findings

Enhanced segmentation accuracy on multiple datasets

Reduced computational complexity

Improved interpretability of learned features

Abstract

Cross-modal medical image segmentation presents a significant challenge, as different imaging modalities produce images with varying resolutions, contrasts, and appearances of anatomical structures. We introduce compositionality as an inductive bias in a cross-modal segmentation network to improve segmentation performance and interpretability while reducing complexity. The proposed network is an end-to-end cross-modal segmentation framework that enforces compositionality on the learned representations using learnable von Mises-Fisher kernels. These kernels facilitate content-style disentanglement in the learned representations, resulting in compositional content representations that are inherently interpretable and effectively disentangle different anatomical structures. The experimental results demonstrate enhanced segmentation performance and reduced computational costs on multiple medical datasets. Additionally, we demonstrate the interpretability of the learned compositional features. Code and checkpoints will be publicly available at: https://github.com/Trustworthy-AI-UU-NKI/Cross-Modal-Segmentation.