Feature selection and classification methodology for the detection of knee-joint disorders

Area of Science:

• Biomedical Engineering
• Signal Processing
• Medical Diagnostics

Background:

• Vibroarthographic (VAG) signals offer early detection for knee joint disorders.
• The nonstationary and nonlinear characteristics of VAG signals necessitate advanced feature extraction techniques.

Purpose of the Study:

• To propose a wavelet-based decomposition method for VAG signals.
• To extract nonlinear features and apply feature selection techniques for improved diagnostic accuracy.
• To evaluate the performance of Least Square Support Vector Machines (LS-SVM) and Random Forest classifiers.

Main Methods:

• VAG signals were decomposed into sub-band signals using wavelet decomposition.
• Nonlinear features including Recurrence Quantification Analysis (RQA), Approximate Entropy (ApEn), and Sample Entropy (SampEn) were extracted.
• Apriori and Genetic Algorithms (GA) were employed for feature selection, identifying the most significant features.

Main Results:

• Feature selection significantly enhanced classification accuracy.
• LS-SVM with the Apriori algorithm achieved the highest accuracy (94.31%) with a False Discovery Rate (FDR) of 0.0892.
• The proposed method demonstrated superior performance compared to previous studies (88% accuracy).

Conclusions:

• The proposed wavelet-based methodology provides an effective non-invasive diagnostic tool for knee joint disorders.
• This approach enhances the ability to distinguish between normal and abnormal VAG signals, improving diagnostic capabilities.