Identifying key gait features associated with the radiological grade of knee osteoarthritis

Abstract

Purpose: Knee osteoarthritis (KOA) is characterized by pain and decreased gait function. This study assessed key features that can be used as mechanical biomarkers for KOA severity and progression. The identified features were validated statistically and were further examined by developing a classification model based on a machine-learning algorithm.

Methods: The study included 227 volunteers with various grades of KOA. The severity of KOA was graded using the Kellgren-Lawrence (KL) system. A total of 165 features were extracted from the gait data. The key features were selected using neighborhood component analysis. The selected features were validated using the t-test. Then, the features were examined by building a classification model using a random forest algorithm.

Results: Twenty features were identified that could discriminate the grade of KOA, including nine features extracted from the knee joint, seven from the hip, two from the ankle and two from the spatiotemporal gait parameters. The t-test showed that some features differed significantly between health and sever group, while some were significantly different among the severe group, and others were significantly different for all KL grades. The areas under the receiver operating characteristic curves for classification were 0.974, 0.992, 0.845, 0.894, and 0.905 for KL grades 0 through 4, respectively.

Conclusion: Key gait features reflecting the grade of KOA were identified. The results of the statistical analysis and machine-learning algorithm show that the features can discriminate the severity of disease according to the KL grade.

Keywords: Cross-sectional study; Diagnostic level II; Gait analysis; Machine learning.