Multi-modal Liver Segmentation and Fibrosis Staging Using Real-world MRI Images

TL;DR

The paper presents a real-world, multi-modal MRI pipeline for non-invasive liver fibrosis assessment, integrating multi-modal MRI co-registration, 3D nnUNet liver segmentation, and STAD-feature-based fibrosis classification via Random Forests. It demonstrates top-tier performance on CARE-Liver 2025 tasks (LiSeg and LiFS) using limited GED4 annotations and multi-center data, with careful data augmentation and registration to enable cross-modality learning. The approach highlights both strong generalization across modalities and challenges from out-of-distribution domain shifts, informing future improvements in registration and feature robustness. Overall, it offers a rapid, reproducible framework for MRI-based liver fibrosis staging that can support earlier diagnosis and clinical decision-making in real-world settings.

Abstract

Liver fibrosis represents the accumulation of excessive extracellular matrix caused by sustained hepatic injury. It disrupts normal lobular architecture and function, increasing the chances of cirrhosis and liver failure. Precise staging of fibrosis for early diagnosis and intervention is often invasive, which carries risks and complications. To address this challenge, recent advances in artificial intelligence-based liver segmentation and fibrosis staging offer a non-invasive alternative. As a result, the CARE 2025 Challenge aimed for automated methods to quantify and analyse liver fibrosis in real-world scenarios, using multi-centre, multi-modal, and multi-phase MRI data. This challenge included tasks of precise liver segmentation (LiSeg) and fibrosis staging (LiFS). In this study, we developed an automated pipeline for both tasks across all the provided MRI modalities. This pipeline integrates pseudo-labelling based on multi-modal co-registration, liver segmentation using deep neural networks, and liver fibrosis staging based on shape, textural, appearance, and directional (STAD) features derived from segmentation masks and MRI images. By solely using the released data with limited annotations, our proposed pipeline demonstrated excellent generalisability for all MRI modalities, achieving top-tier performance across all competition subtasks. This approach provides a rapid and reproducible framework for quantitative MRI-based liver fibrosis assessment, supporting early diagnosis and clinical decision-making. Code is available at https://github.com/YangForever/care2025_liver_biodreamer.

Figures (4)

• Figure 1: The automated pipeline for liver segmentation and fibrosis staging. A 3D nnUNet was trained on the dataset involving the original annotated GED4 data, augmented GED4 data, and the registered multi-modality data. After that, the fibrosis staging was performed by the Random Forest classifiers based on the STAD (Shape, Texture, Appearance, Directional) features selected from the segmentation predictions.
• Figure 2: Quantitative evaluation of the method on LiSeg and LiFS. (A) shows the Dice and Hausdorff distance (HD) of different model configurations on (Ai.) the Training dataset and (Aii.)(Aiii.) the hold-out Validation dataset. (Aiv.)(Av.) illustrates the Model-All-Norm model performance on the hold-out Test dataset. (B) indicates the AUC and ACC of the Model-All on (Bi.)(Bii.)the hold-out Validation dataset and (Biii.)(Biv.) the hold-out Test dataset. (N/S denotes Not Submitted to the scoring system, and N/A is Not Applicable).
• Figure 3: Results of co-registration, pseudo-labelling, and corresponding predictions. All the modalities are from a randomly selected case with the 2D slices taken from the middle of the MRI volume along the z-axis.
• Figure 4: Importance analysis of the features selected for Random Forest classifiers. Standard deviation is shown as the error bar in black.