Automatic Recognition of Gait Patterns Exhibiting Patellofemoral Pain Syndrome Using a Support Vector Machine Approach

• Published in: IEEE journal of biomedical and health informatics 2009-09, Vol.13 (5), p.810-817
• Main Authors: Lai, D.T.H., Levinger, P., Begg, R.K., Gilleard, W.L., Palaniswami, M.
• Format: Article
• Language: English
• Subjects: Adolescent; Adult; Algorithms; Artificial Intelligence; Female; Foot; Force measurement; Gait - physiology; Gait analysis; ground reaction forces (GRFs); Humans; Kinematics; Knee; Middle Aged; Pain; patellofemoral pain syndrome (PFPS); Patellofemoral Pain Syndrome - diagnosis; Patellofemoral Pain Syndrome - physiopathology; Pathology; Pattern recognition; Pattern Recognition, Automated - methods; Support vector machine classification; Support vector machines; support vector machines (SVMs); Testing; Tibia - physiology
• ISSN: 1089-7771; 2168-2194; 1558-0032; 2168-2208
• DOI: 10.1109/TITB.2009.2022927

Patellofemoral pain syndrome (PFPS) is a common disorder that afflicts people across all age groups, and results in various degrees of knee pain. The diagnosis of PFPS is difficult since the exact biomechanical factors and the extent to which they are affected by the disorder are still unknown. Recent research has reported significant statistical differences in ground reaction forces (GRFs) and foot kinematics, which could be indicative of PFPS, but the interrelationship between many of these measures and the pathology have been absent so far. In this paper, we applied the support vector machines (SVMs) to detect PFPS gait based on 14 GRF and 16 foot kinematic features recorded from 27 subjects (14 healthy and 13 with PFPS). The influence of combined gait parameters on classification performance was investigated through the use of a feature-selection algorithm. The optimal feature set was then compared against the most statistically significant individual features ( p < 0.05) found by previous study. Test results indicated that GRF features alone resulted in a higher leave-one-out (LOO) classification accuracy (85.15%) compared to 74.07% using only kinematic features. A hill-climbing feature-selection algorithm was applied to determine the subset of combined kinematic and kinetic features, which provided optimal classifier performance. This subset, which consists of six features (two from GRF and four from foot kinematic features), provided an improved LOO accuracy of 88.89% . The optimal feature set detected by the SVM, which best identified gait characteristics of PFPS, was found to be closely related to inferential statistical analysis with the added distinction that the SVM could potentially be deployed as an automated system for detecting gait changes in patients with PFPS.