Distinguishing ACL patients from healthy individuals using multilayer perceptron on motion patterns

Anterior Cruciate Ligament (ACL) injury is an injury of knee joints happened to many athletes. It has the significant impact on the patients' movement in their daily life and sport activities. Thus, it is important to detect the ACL injury in an early stage. So that, the proper treatment can be operated in time. This paper proposes a novel method to seek out differences of gait patterns between the ACL reconstructed patients and the healthy individuals, along with classifying the class where the gait data belongs to. The multilayer perceptron (MLP) will then be applied as the classification tool. In the experiment, 8 subjects are used to validate the proposed method. Each subject contains nine different observed variables of gait information gathered from the 3D motion-capture camera system. The proposed method can achieve a very promising performance of about 90% accuracy. Also, the weights gaining from the MLP model with respect to each particular gait variable are plotted, in order to determine the key variables that are significant to discriminate the ACL reconstructed patients from the healthy individuals. The detected three dominant variables are the ground reaction force, the ankle joint moment, and the ankle joint angle.