Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Leveraging Gait Patterns as Biomarkers: An attention-guided Deep Multiple Instance Learning Network for Scoliosis Classification

Haiqing Li, Yuzhi Guo, Feng Jiang, Qifeng Zhou, Hehuan Ma, Junzhou Huang

TL;DR

The paper tackles noninvasive, scalable scoliosis screening by leveraging gait patterns as biomarkers. It introduces Gait-MIL, an attention-guided deep multi-instance learning framework that clusters gait frames into phase-based bags, applies intra-bag attention, and aggregates bag features with inter-bag attention, trained with a triplet and cross-entropy loss. On the Scoliosis1K dataset, Gait-MIL achieves $84.8\%$ accuracy, $99.0\%$ sensitivity, and $79.6\%$ specificity, notably improving Neutral-case discrimination and robustness to class imbalance. This approach enables accurate, large-scale screening in settings where X-ray-based methods are impractical, with potential impact on early intervention and public health.

Abstract

Scoliosis is a spinal curvature disorder that is difficult to detect early and can compress the chest cavity, impacting respiratory function and cardiac health. Especially for adolescents, delayed detection and treatment result in worsening compression. Traditional scoliosis detection methods heavily rely on clinical expertise, and X-ray imaging poses radiation risks, limiting large-scale early screening. We propose an Attention-Guided Deep Multi-Instance Learning method (Gait-MIL) to effectively capture discriminative features from gait patterns, which is inspired by ScoNet-MT's pioneering use of gait patterns for scoliosis detection. We evaluate our method on the first large-scale dataset based on gait patterns for scoliosis classification. The results demonstrate that our study improves the performance of using gait as a biomarker for scoliosis detection, significantly enhances detection accuracy for the particularly challenging Neutral cases, where subtle indicators are often overlooked. Our Gait-MIL also performs robustly in imbalanced scenarios, making it a promising tool for large-scale scoliosis screening.

Leveraging Gait Patterns as Biomarkers: An attention-guided Deep Multiple Instance Learning Network for Scoliosis Classification

TL;DR

The paper tackles noninvasive, scalable scoliosis screening by leveraging gait patterns as biomarkers. It introduces Gait-MIL, an attention-guided deep multi-instance learning framework that clusters gait frames into phase-based bags, applies intra-bag attention, and aggregates bag features with inter-bag attention, trained with a triplet and cross-entropy loss. On the Scoliosis1K dataset, Gait-MIL achieves accuracy, sensitivity, and specificity, notably improving Neutral-case discrimination and robustness to class imbalance. This approach enables accurate, large-scale screening in settings where X-ray-based methods are impractical, with potential impact on early intervention and public health.

Abstract

Scoliosis is a spinal curvature disorder that is difficult to detect early and can compress the chest cavity, impacting respiratory function and cardiac health. Especially for adolescents, delayed detection and treatment result in worsening compression. Traditional scoliosis detection methods heavily rely on clinical expertise, and X-ray imaging poses radiation risks, limiting large-scale early screening. We propose an Attention-Guided Deep Multi-Instance Learning method (Gait-MIL) to effectively capture discriminative features from gait patterns, which is inspired by ScoNet-MT's pioneering use of gait patterns for scoliosis detection. We evaluate our method on the first large-scale dataset based on gait patterns for scoliosis classification. The results demonstrate that our study improves the performance of using gait as a biomarker for scoliosis detection, significantly enhances detection accuracy for the particularly challenging Neutral cases, where subtle indicators are often overlooked. Our Gait-MIL also performs robustly in imbalanced scenarios, making it a promising tool for large-scale scoliosis screening.

Paper Structure

This paper contains 16 sections, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Frame-Level Clustering
  4. Attention-Based Pooling
  5. Bag-Level Feature Aggregation
  6. Gait-Based Scoliosis Classification
  7. Experiments
  8. Dataset
  9. Evaluation Protocol
  10. Implementation Details
  11. Results
  12. Comparison with State-of-the-Art
  13. Ablation Studies
  14. Conclusion
  15. ACKNOWLEDGMENTS
  16. ...and 1 more sections

Figures (3)

  • Figure 1: Illustrations of Cobb angle calculation, used to evaluate scoliosis severity.
  • Figure 2: The overall flowchart of the proposed Gait-MIL. It consists of four parts: (1) K-Means is used to cluster the sampled gait frames into distinct bags, enabling the model to extract specific features from different gait phases, (2) Attention-based pooling is applied to focus on the most informative frames within each bag, and (3) An attention mechanism is used to aggregate bag features into a global representation, (4) The triplet and cross-entropy losses are utilized to drive the training process.
  • Figure 3: Comparison of Confusion Matrices for Gait-MIL and ScoNet-MT.