New Gait Representation Maps for Enhanced Recognition in Clinical Gait Analysis

Area of Science:

• Biomedical Engineering
• Computer Vision
• Clinical Biomechanics

Background:

• Gait analysis is crucial for evaluating neuromuscular and musculoskeletal disorders.
• Traditional visual gait assessment is subjective and inconsistent.
• Automated gait analysis using Gait Energy Images (GEI) lacks detailed spatial and temporal information.

Purpose of the Study:

• To introduce novel gait representations that capture detailed spatial, temporal, and boundary information.
• To overcome the limitations of conventional GEI in automated gait analysis.
• To improve the accuracy and reliability of clinical gait assessments.

Main Methods:

• Developed four novel gait representations: tGBI, cGEI, tGDI, and cBIT.
• Constructed representations from binary silhouette sequences using transformations preserving structural and dynamic information.
• Evaluated representations on the INIT GAIT dataset using machine learning models for gait impairment classification.

Main Results:

• Proposed gait representations consistently outperformed conventional GEI.
• Improvements were observed across multiple machine learning models and classification tasks (4 and 6 classes).
• The novel methods demonstrated enhanced capture of critical gait cycle features.

Conclusions:

• The novel gait representations (tGBI, cGEI, tGDI, cBIT) significantly enhance automated clinical gait analysis.
• These methods offer a more accurate and reliable approach compared to traditional GEI.
• The findings support the potential for improved diagnosis and monitoring of gait impairments.