Reducing Textural Bias Improves Robustness of Deep Segmentation Models

TL;DR

This paper investigates how reducing textural bias in deep segmentation models enhances their robustness to domain shifts in 3D medical imaging, using extensive experiments with simulated noise on MRI data.

Contribution

It provides a comprehensive empirical study demonstrating that training with simulated textural noise improves model robustness and texture invariance in medical image segmentation.

Findings

Applying specific textural noise during training enhances robustness to unseen noise.

Texture invariant models perform better on corrupted scans.

Extensive 176 experiments validate the approach.

Abstract

Despite advances in deep learning, robustness under domain shift remains a major bottleneck in medical imaging settings. Findings on natural images suggest that deep neural models can show a strong textural bias when carrying out image classification tasks. In this thorough empirical study, we draw inspiration from findings on natural images and investigate ways in which addressing the textural bias phenomenon could bring up the robustness of deep segmentation models when applied to three-dimensional (3D) medical data. To achieve this, publicly available MRI scans from the Developing Human Connectome Project are used to study ways in which simulating textural noise can help train robust models in a complex semantic segmentation task. We contribute an extensive empirical investigation consisting of 176 experiments and illustrate how applying specific types of simulated textural noise prior to training can lead to texture invariant models, resulting in improved robustness when segmenting scans corrupted by previously unseen noise types and levels.