Pose as Clinical Prior: Learning Dual Representations for Scoliosis Screening

2025-08-31

AI Summary

Existing AI-based scoliosis screening methods predominantly rely on contour images, overlooking clinically critical postural asymmetries and suffering from a lack of large-scale, expert-annotated pose data. Moreover, raw 2D keypoint estimations are highly sensitive to noise and inherently discrete, hindering robust modeling of subtle asymmetries. To address these limitations, we propose the Dual-Representation Framework (DRF): (1) we introduce Scoliosis1K-Pose—the first large-scale, expert-annotated 2D pose dataset comprising over 1,000 scoliosis cases; (2) we design a continuous skeleton graph encoding alongside a discrete Postural Asymmetry Vector (PAV), which encodes clinically grounded asymmetry patterns; and (3) we integrate PAV as a clinical prior to guide an attention mechanism that focuses on diagnostically relevant anatomical regions, enabling end-to-end multimodal pose representation learning. Experiments demonstrate state-of-the-art performance across multiple metrics; visualizations confirm accurate capture of clinically meaningful asymmetry patterns, significantly improving both screening accuracy and interpretability.

Technology Category

• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning
• Computer Vision: Scene Analysis & Understanding
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Recent AI-based scoliosis screening methods primarily rely on large-scale silhouette datasets, often neglecting clinically relevant postural asymmetries-key indicators in traditional screening. In contrast, pose data provide an intuitive skeletal representation, enhancing clinical interpretability across various medical applications. However, pose-based scoliosis screening remains underexplored due to two main challenges: (1) the scarcity of large-scale, annotated pose datasets; and (2) the discrete and noise-sensitive nature of raw pose coordinates, which hinders the modeling of subtle asymmetries. To address these limitations, we introduce Scoliosis1K-Pose, a 2D human pose annotation set that extends the original Scoliosis1K dataset, comprising 447,900 frames of 2D keypoints from 1,050 adolescents. Building on this dataset, we introduce the Dual Representation Framework (DRF), which integrates a continuous skeleton map to preserve spatial structure with a discrete Postural Asymmetry Vector (PAV) that encodes clinically relevant asymmetry descriptors. A novel PAV-Guided Attention (PGA) module further uses the PAV as clinical prior to direct feature extraction from the skeleton map, focusing on clinically meaningful asymmetries. Extensive experiments demonstrate that DRF achieves state-of-the-art performance. Visualizations further confirm that the model leverages clinical asymmetry cues to guide feature extraction and promote synergy between its dual representations. The dataset and code are publicly available at https://zhouzi180.github.io/Scoliosis1K/.

Problem

• Addressing scarcity of annotated pose datasets for scoliosis screening
• Overcoming noise sensitivity in raw pose coordinates modeling
• Integrating clinical asymmetry indicators into AI screening methods

Innovation

• Extended Scoliosis1K dataset with 2D pose annotations
• Dual Representation Framework combining skeleton maps and asymmetry vectors
• PAV-Guided Attention module using clinical prior for feature extraction