The outcomes of a learned classification model can be used to determine functional status post total knee arthroplasty (tka) based on wearable sensor data

In a clinical setting, there is a need for simple gait kinematic measurements to facilitate objective unobtrusive patient monitoring. The objective of this study is to determine if a learned classification model's output can be used to monitor a person's recovery status post-TKA. The gait kinematics of 20 asymptomatic and 17 people with TKA were measured using a full-body Xsens model 1 . The experimental group was measured at 6 weeks, 3, 6, and 12 months post-surgery. Joint angles of the ankle, knee, hip, and spine per stride (10 strides) were extracted from the Xsens software (MVN Awinda studio 4.4) 1 . Statistical features for each subject at each evaluation moment were derived from the kinematic time-series data. We normalised the features using standard scaling 2 . We trained a logistic regression (LR) model using L1-regularisation on the 6 weeks post-surgery data2–4. After training, we applied the trained LR- model to the normalised features computed for the subsequent timepoints. The model returns a score between 0 (100% confident the person is an asymptomatic control) and 1 (100% confident this person is a patient). The decision boundary is set at 0.5. The classification accuracy of our LR-model was 94.58%. Our population's probability of belonging to the patient class decreases over time. At 12 months post-TKA, 38% of our patients were classified as asymptomatic.