One Sequence to Segment Them All: Efficient Data Augmentation for CT and MRI Cross-Domain 3D Spine Segmentation

2026-05-04

AI Summary

This work addresses the limited generalization of models trained on single-sequence data across cross-modality (CT/MRI) and cross-contrast medical images, as well as performance degradation due to scarce annotated data. To this end, the authors propose an efficient data augmentation strategy tailored for cross-modality transfer. Their approach is the first to systematically quantify the impact of data augmentation on cross-modality generalization and introduces a GPU-optimized augmentation pipeline. Using only single-sequence training data, the method substantially improves model robustness for spine segmentation on unseen CT/MRI sequences. Experiments demonstrate an average 155% increase in Dice coefficient across seven out-of-domain datasets, with negligible in-domain performance loss (average drop of merely 0.008%) and approximately 10% higher training efficiency. The accompanying toolbox is open-sourced and compatible with mainstream frameworks such as nnUNet and MONAI.

Abstract

Deep learning-based medical image segmentation is increasingly used to support clinical diagnosis and develop new treatment strategies. However, model performance remains limited by the scarcity of high-quality annotated data and insufficient generalization across imaging protocols. This limitation is particularly evident in MRI and CT, where models are typically trained on a single acquisition sequence and exhibit reduced robustness when applied to unseen sequences or contrasts. Although data augmentation is widely used to improve general robustness on medical images, its impact on cross-modality generalization has not been quantitatively explored. In this work, we study a targeted set of data augmentation techniques designed to improve cross-modality transfer. We train three spine segmentation models, each on a single-modality/sequence dataset, and evaluate them across seven out-of-distribution datasets (spanning CT and MRI), reflecting a realistic single-sequence training and multi-sequence/contrast/modality deployment scenario. Our results demonstrate substantial performance gains on unseen domains (average Dice gain of 155 %) while preserving in-domain accuracy (average Dice decrease of 0.008 %), including effective transfer between CT and MRI. To mitigate the computational cost typically associated with strong data augmentation, we implement GPU-optimized augmentations that maintain, and even improve, training efficiency by approximately 10 %. We release our approach as an open-source toolbox, enabling seamless integration into commonly used frameworks such as nnUNet and MONAI. These augmentations significantly enhance robustness to heterogeneous clinical imaging scenarios without compromising training speed.

Problem

• cross-modality
• data augmentation
• medical image segmentation
• domain generalization
• spine segmentation

Innovation

• cross-modality segmentation
• data augmentation
• spine segmentation
• domain generalization
• GPU-accelerated augmentation