Predicting Severe Knee Arthritis Based on Two Inertial Measurement Unit Sensors as a Dynamic Coordinate System Using Classical Machine Learning.

Abstract

Background: Aging of societies in recent and upcoming years has made musculoskeletal disorders a significant challenge for healthcare system. Knee osteoarthritis (KOA) is a progressive musculoskeletal disorder that is typically diagnosed using radiographs. Considering the drawbacks of X-ray imaging, such as exposure to ionizing radiation, the need for a noninvasive, low-cost alternative method for diagnosing KOA is essential. The purpose of this study was to evaluate the ability of a wearable device to differentiate between healthy individuals and those with severe osteoarthritis (grade 4).

Methods: The wearable device consisted of two inertial measurement unit (IMU) sensors, one on the lower leg and one on the thigh. One of the sensors is used as a dynamic coordinate system to improve the accuracy of the measurements. In this study, to discriminate between 1433 labeled IMU signals collected from 15 healthy individuals and 15 people with severe KOA aged over 45, new features were extracted and defined in dynamic coordinates. These features were employed in four different classifiers: (1) naive Bayes, (2) K-nearest neighbors (KNNs), (3) support vector machine, and (4) random forest. Each classifier was evaluated using the 10-fold cross-validation method (K = 10). The data were applied to these models, and based on their outputs, four performance metrics - accuracy, precision, sensitivity, and specificity - were calculated to assess the classification of these two groups using the mentioned software.

Results: The evaluation of the selected classifiers involved calculating the four specified metrics and their average and variance values. The highest accuracy was achieved by KNN, with an accuracy of 93.71 ± 1.1 and a precision of 93 ± 1.31.

Conclusion: The novel features based on the dynamic coordinate system, along with the success of the proposed KNN model, demonstrate the effectiveness of the proposed algorithm in diagnosing between signals received from healthy individuals and patients. The proposed algorithm outperforms existing methods in similar articles in sensitivity showing an improvement of 4% and at least. The main objective of this study is to investigate the feasibility of using a wearable device as an auxiliary tool in the diagnosis of arthritis. The reported results in this study are related to two groups of individuals with severe arthritis (grade 4), and there is a possibility of weaker results with the current method.