Probability-weighted ensemble support vector machine for intelligent recognition of moving wear debris from joint implant

Abstract

Friction-induced wear debris from joint implants are effective resources in investigating artificial joint wear and cellular immune response mechanisms. To improve the accuracy of wear debris analysis, an intelligent recognition method is developed for wear debris measurement under motion conditions. In this method, the multi-view image sequence of moving wear debris is captured to acquire the variations of aspect ratio, area, and roundness features. Then multiple SVM models are integrated to identify wear debris types based on weighted probability to improve the accuracy. The proposed method can achieve a classification accuracy of 90.51%, which is better than HIVE-COTE2.0, MultiRocket, and other time series classification algorithms. This method can be applied to monitor wear status of artificial joint articulating surfaces.