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

2026-07-22 · Source: Computer Vision and Pattern Recognition · Field: Science & Research — Health & Medical Research, Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

A systematic benchmark compared seven intensity normalization methods for 3D knee MRI meniscus segmentation using a 3D U-Net model. Researchers trained models on the IWOAI 2019 dataset. Evaluation occurred on both internal and external test sets, specifically SKM-TEA, to assess generalizability. The methods included standard scaling, histogram-based techniques, and a Gaussian Mixture Model (GMM)-based approach. While internal performance was similar across methods, Z-score, Nyúl histogram matching, and CLAHE demonstrated greater robustness on external data. However, the observed differences between normalization methods were small compared to the significant performance drop attributed to domain shift between the IWOAI 2019 and SKM-TEA datasets. This highlights that while intensity normalization measurably affects model generalizability, its impact is limited relative to domain shift, necessitating complementary strategies for robust clinical deployment.

Key takeaway

For Machine Learning Engineers deploying medical imaging models, prioritize Z-score, Nyúl histogram matching, or CLAHE for 3D knee MRI intensity normalization. If you are addressing cross-scanner variability, recognize these methods offer only marginal gains compared to domain shift. You must integrate additional domain adaptation techniques to achieve truly robust out-of-the-box clinical performance.

Key insights

Intensity normalization improves MRI segmentation generalizability, but domain shift remains a larger challenge for clinical deployment.

Principles

• Z-score, Nyúl, and CLAHE show robustness.
• Domain shift impacts generalizability more.
• Complementary strategies are needed.

Method

A 3D U-Net model was trained on IWOAI 2019, then evaluated on internal and SKM-TEA external data, comparing seven intensity normalization methods.

In practice

• Prioritize Z-score, Nyúl, or CLAHE for MRI.
• Expect limited gains from normalization alone.
• Combine normalization with domain adaptation.

Topics

• 3D Knee MRI
• Meniscus Segmentation
• Intensity Normalization
• U-Net Model
• Domain Shift
• Cross-Domain Generalizability

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• A Systematic Benchmark of Intensity Normalisation Methods for 3D Knee MRI Segmentation and Cross-Domain Generalisability — Intensity normalisation minimally improves 3D knee MRI segmentation generalisability; domain shift remains the primary performance challenge.
• ROBUST-WT: Robust Uncertainty-aware Segmentation Transform via Whitening and Training Enhancements — ROBUST-WT demonstrates that targeted training enhancements can substantially improve medical image segmentation robustness without altering core architecture.
• Domain Shift in Echocardiography: Interpretable Quantification and Prediction of Cross-Dataset Left Ventricular Segmentation — Echocardiographic domain shift is measurable and its impact on LV segmentation can be reduced by geometry-aware preprocessing and predicted using representation-specific metrics.
• MED-DSLC: Multi-Expert-Domain Classification via Domain Supervision and Logit Calibration — Logit miscalibration in merged VLM experts degrades multi-domain performance; MED-DSLC restores global comparability through domain supervision and logit scaling.
• MMBU: A Massive Multi-modal Biomedical Understanding Benchmark to Probe the Perception Capabilities of Vision-Language Models — The MMBU benchmark reveals current VLMs lack robust visual perception and domain generalization in biomedical contexts.
• IEEE Transactions on Neural Networks and Learning Systems, Volume 37, Issue 3, March 2026 — The issue highlights advancements across neural networks, computer vision, and machine learning applications.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.