The impact of feature extraction and selection for the classification of gait patterns between ACL deficient and intact knees based on different classification models

Abstract

AbstractThe anterior cruciate ligament (ACL) plays an important role in stabilizing translation and rotation of the tibia relative to the femur. Individuals with ACL deficiency usually demonstrate alterations in gait characteristics. Evidence indicates that walking speed, alterations in kinetics and kinematics on the ACL deficient limb, and inter-limb asymmetries between deficient and intact knees may contribute to poor long-term outcomes following ACL deficiency. They corrode function of the knee joint and put it at higher risk of degeneration. For the purpose of developing an automatic and highly accurate system for detection of ACL deficiency, this study investigated the classification capability of different dynamical features extracted from gait kinematic and kinetic signals when evaluating their impact on different classification models. A general feature extraction framework was proposed and various dynamical features, such as recurrence rate, determinism and entropy from the recurrence quantification analysis, fuzzy entropy, Teager-Kaiser energy feature and statistical analysis, were included. Different classification models, including support vector machine (SVM), K-nearest neighbor (KNN), naive Bayes (NB) classifier, decision tree (DT) classifier and ensemble learning based Adaboost (ELA) classifier, derived for discriminant analysis of multiple dynamical gait features were evaluated for a comparative study. The effectiveness of this strategy was verified using a dataset of knee, hip and ankle kinematic and kinetic waveforms from 43 patients with unilateral ACL deficiency. When evaluated with 2-fold, 10-fold and leave-one-out cross-validation styles, the highest classification accuracy for discriminating between groups of ACL deficient and contralateral ACL intact knees was reported to be 91.22 $$\%$$ % , 95.12$$\%$$ % and 96.34$$\%$$ % , respectively,by using the SVM classifier and the optimal feature set. For other four classifiers, KNN achieved the accuracy of 78.05$$\%$$ % , 85.37$$\%$$ % and 87.80$$\%$$ % , respectively. NB achieved the accuracy of 57.56$$\%$$ % , 60.98$$\%$$ % and 61.22$$\%$$ % , respectively. DT achieved the accuracy of 77.56$$\%$$ % , 80.49$$\%$$ % and 83.66$$\%$$ % , respectively. ELA achieved the accuracy of 73.66$$\%$$ % , 78.05$$\%$$ % and 79.27$$\%$$ % , respectively. Compared with other state-of-the-art methods, the results demonstrate superior performance and support the validity of the proposed method.