A Two-stream Convolutional Network for Musculoskeletal and Neurological Disorders Prediction

Area of Science:

• Biomechanics
• Medical Imaging and Data Analysis
• Artificial Intelligence in Healthcare

Background:

• Musculoskeletal and neurological disorders frequently cause mobility issues in the elderly, impacting their quality of life.
• Manual analysis of walking data is time-consuming, requires expertise, and can be subjective.
• Automated analysis using deep learning shows potential for early diagnosis by identifying complex patterns.

Purpose of the Study:

• To develop an automated system for analyzing walking problems in older adults.
• To improve diagnostic accuracy by incorporating both individual joint and inter-joint movement features.
• To overcome limitations of existing deep learning methods that focus solely on individual joint data.

Main Methods:

• A novel two-stream deep learning framework was proposed, processing joint position and relative joint displacement time series separately.
• A mid-layer fusion module was developed to integrate features from both streams for enhanced pattern recognition.
• The system was validated using a 3D skeleton motion dataset from 45 patients with mobility-affecting disorders.

Main Results:

• The proposed two-stream framework achieved a prediction accuracy of 95.56% in diagnosing walking problems.
• This accuracy significantly outperforms existing state-of-the-art methods.
• The integration of inter-joint features proved crucial for improved performance.

Conclusions:

• The developed deep learning system effectively analyzes walking motion data for diagnosing musculoskeletal and neurological disorders.
• Explicitly incorporating inter-joint features alongside individual joint data enhances diagnostic prediction.
• This approach offers a promising tool for objective and early detection of mobility impairments in older adults.