Hybridized deep learning goniometry for improved precision in Ehlers-Danlos Syndrome (EDS) evaluation

Area of Science:

• Medical Imaging
• Deep Learning
• Connective Tissue Diseases

Background:

• Generalized Joint Hypermobility (GJH) is crucial for diagnosing Ehlers-Danlos Syndrome (EDS).
• EDS presents complex symptoms that can mimic other conditions.
• Current diagnostic methods for joint mobility can be inconsistent.

Purpose of the Study:

• To develop HybridPoseNet, a novel image-based goniometry system for precise joint angle measurement in EDS assessments.
• To enhance diagnostic accuracy for connective tissue disorders using advanced deep learning.

Main Methods:

• HybridPoseNet integrates Convolutional Neural Networks (CNNs, specifically MobileNet-V2) for spatial pattern recognition and HyperLSTM units for sequential data processing.
• The system was trained on diverse datasets and fine-tuned using video data from 50 individuals with EDS, focusing on hyperextendable joints.
• Model performance was validated against manual goniometry measurements using Spearman's correlation and human joint position labeling.

Main Results:

• HybridPoseNet demonstrated high correlation with manual measurements: thumb (rho=0.847), elbows (rho=0.822), knees (rho=0.839), and fifth fingers (rho=0.896).
• The model showed consistent performance across all joint assessments, offering a unified approach.
• An approximate 20% increase in accuracy was observed compared to standard pose estimation libraries.

Conclusions:

• HybridPoseNet offers a valuable, accurate, and consistent method for assessing joint mobility in the context of EDS.
• This image-based goniometry system represents a significant advancement in the medical diagnostics of connective tissue diseases.
• The system's normalized approach enhances the understanding and review of joint mobility.