Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MeniMV: A Multi-view Benchmark for Meniscus Injury Severity Grading

Shurui Xu, Siqi Yang, Jiapin Ren, Zhong Cao, Hongwei Yang, Mengzhen Fan, Yuyu Sun, Shuyan Li

TL;DR

The paper introduces MeniMV, a large-scale, dual-view knee MRI dataset with horn-specific four-level severity annotations (0–3) for anterior and posterior meniscal horns, collected from three medical centers and validated by clinicians. It formulates horn-specific grading as a dual-head classification task using fused sagittal and coronal features extracted by various backbones, including state-of-the-art Transformers, and trains with a hybrid loss that combines focal imbalance handling and cross-view alignment. Experimental results show Transformer-based encoders, particularly Swin-UNETR with pretraining, provide the strongest performance, while concatenation-based fusion outperforms other fusion strategies; cross-center robustness is best with pretrained transformers, though demographic gaps remain. Overall, MeniMV establishes a high-quality benchmark for automated, clinically relevant grading of meniscal horn injuries and lays groundwork for future integration with clinical metadata and extension to other joint structures.

Abstract

Precise grading of meniscal horn tears is critical in knee injury diagnosis but remains underexplored in automated MRI analysis. Existing methods often rely on coarse study-level labels or binary classification, lacking localization and severity information. In this paper, we introduce MeniMV, a multi-view benchmark dataset specifically designed for horn-specific meniscus injury grading. MeniMV comprises 3,000 annotated knee MRI exams from 750 patients across three medical centers, providing 6,000 co-registered sagittal and coronal images. Each exam is meticulously annotated with four-tier (grade 0-3) severity labels for both anterior and posterior meniscal horns, verified by chief orthopedic physicians. Notably, MeniMV offers more than double the pathology-labeled data volume of prior datasets while uniquely capturing the dual-view diagnostic context essential in clinical practice. To demonstrate the utility of MeniMV, we benchmark multiple state-of-the-art CNN and Transformer-based models. Our extensive experiments establish strong baselines and highlight challenges in severity grading, providing a valuable foundation for future research in automated musculoskeletal imaging.

MeniMV: A Multi-view Benchmark for Meniscus Injury Severity Grading

TL;DR

The paper introduces MeniMV, a large-scale, dual-view knee MRI dataset with horn-specific four-level severity annotations (0–3) for anterior and posterior meniscal horns, collected from three medical centers and validated by clinicians. It formulates horn-specific grading as a dual-head classification task using fused sagittal and coronal features extracted by various backbones, including state-of-the-art Transformers, and trains with a hybrid loss that combines focal imbalance handling and cross-view alignment. Experimental results show Transformer-based encoders, particularly Swin-UNETR with pretraining, provide the strongest performance, while concatenation-based fusion outperforms other fusion strategies; cross-center robustness is best with pretrained transformers, though demographic gaps remain. Overall, MeniMV establishes a high-quality benchmark for automated, clinically relevant grading of meniscal horn injuries and lays groundwork for future integration with clinical metadata and extension to other joint structures.

Abstract

Precise grading of meniscal horn tears is critical in knee injury diagnosis but remains underexplored in automated MRI analysis. Existing methods often rely on coarse study-level labels or binary classification, lacking localization and severity information. In this paper, we introduce MeniMV, a multi-view benchmark dataset specifically designed for horn-specific meniscus injury grading. MeniMV comprises 3,000 annotated knee MRI exams from 750 patients across three medical centers, providing 6,000 co-registered sagittal and coronal images. Each exam is meticulously annotated with four-tier (grade 0-3) severity labels for both anterior and posterior meniscal horns, verified by chief orthopedic physicians. Notably, MeniMV offers more than double the pathology-labeled data volume of prior datasets while uniquely capturing the dual-view diagnostic context essential in clinical practice. To demonstrate the utility of MeniMV, we benchmark multiple state-of-the-art CNN and Transformer-based models. Our extensive experiments establish strong baselines and highlight challenges in severity grading, providing a valuable foundation for future research in automated musculoskeletal imaging.

Paper Structure

This paper contains 15 sections, 5 equations, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. MeniMV Dataset
  3. Data Collection and Labeling
  4. Dataset Properties and Context
  5. Multi-view Meniscus Injury Grading
  6. Task Definition
  7. Multi-view Encoder
  8. Classification Head
  9. Loss Function
  10. Experimental Analysis
  11. Backbone and Method Analysis
  12. Fusion Strategy Analysis
  13. Cross-center Robustness (LOCO)
  14. Demographic Robustness
  15. Conclusion

Figures (2)

  • Figure 1: Example from the MeniMV dataset. Each case includes two views—sagittal and coronal—with anterior and posterior meniscal horns annotated across four injury grades.
  • Figure 2: The selection procedure of MeniMV. The four most diagnostically relevant slices from both sagittal and coronal MRI series are selected by chief orthopedic physicians.