Selection of clinical features for pattern recognition applied to gait analysis

Area of Science:

• Biomedical Engineering
• Medical Informatics
• Biomechanics

Background:

• Gait analysis is crucial for diagnosing various medical conditions.
• Stereophotogrammetric systems provide rich data for gait assessment.
• Identifying relevant gait parameters is essential for accurate classification.

Purpose of the Study:

• To develop and evaluate a feature selection method for gait analysis data.
• To identify the optimal subset of gait parameters for classifying healthy versus diseased subjects.
• To assess the performance of common classification algorithms using selected gait features.

Main Methods:

• Applied stereophotogrammetry to capture gait data.
• Utilized an exhaustive feature selection method to evaluate gait parameter combinations.
• Tested Support Vector Machine (SVM), Naive Bayes, and K-Nearest Neighbor (KNN) classifiers.
• Assessed classification accuracy and the number of selected features.

Main Results:

• A high classification accuracy exceeding 97% was achieved.
• The full set of gait features was found to be redundant.
• Subsets of only 3 to 5 features were sufficient to maintain high accuracy.
• Step length and swing speed were identified as the most informative features.

Conclusions:

• Feature selection significantly reduces dimensionality in gait analysis without compromising accuracy.
• Stereophotogrammetry combined with advanced classification techniques offers a powerful tool for medical diagnosis.
• Minimal feature sets can effectively detect gait anomalies, improving diagnostic efficiency.