A Systematic Benchmark of Intensity Normalisation Methods for 3D Knee MRI Segmentation and Cross-Domain Generalisability

2026-07-22

AI Summary

This study addresses the challenge of intensity inconsistency in MRI scans arising from differences in scanners and acquisition protocols by systematically evaluating seven intensity normalization methods—including Z-score, Nyúl histogram matching, CLAHE, and Gaussian mixture models—on the cross-domain generalization performance of a 3D U-Net for meniscus segmentation in knee MRI. Leveraging the IWOAI 2019 dataset for training and testing on both internal and external datasets such as SKM-TEA, this work presents the first comprehensive comparison of normalization strategies in 3D knee MRI segmentation. Results indicate that Z-score, Nyúl matching, and CLAHE exhibit relatively robust performance; however, the gains conferred by normalization are modest and insufficient to offset the substantial performance degradation caused by inter-dataset distribution shifts, underscoring the necessity of integrating additional robustness mechanisms to enhance generalization.

Abstract

Robust out-of-the-box performance is essential for the clinical deployment of deep learning models in medical imaging. An important but underexplored factor affecting model generalisability is intensity normalisation, particularly for magnetic resonance imaging (MRI), where image intensities vary across scanners and protocols. In this study, we systematically compared seven normalisation methods and their impact on the performance of a 3D U-Net model for meniscus segmentation from knee MRI. The methods included standard scaling approaches, histogram-based techniques, and a Gaussian Mixture Model (GMM)-based method. Models were trained on the IWOAI 2019 dataset and evaluated on both internal and external test sets (SKM-TEA) to assess generalisability. Performance was similar internally but differences were significant on external data, with Z-score, Nyúl histogram matching, and CLAHE showing greater robustness than other methods. However, these differences were small compared to the significant performance drop observed between datasets. Overall, while intensity normalisation had a measurable effect on model generalisability, its impact was limited relative to the effects of domain shift, highlighting the need for complementary strategies for robust deployment.

Problem

• intensity normalisation
• MRI
• cross-domain generalisability
• knee segmentation
• domain shift

Innovation

• intensity normalisation
• cross-domain generalisability
• 3D MRI segmentation
• systematic benchmarking
• domain shift