Machine Learning Model for Classification of Shoulder Implant Manufacturer Using X-Ray Images

Abstract

Recently, synthetic prosthesis built from metals and plastic components are often used to mitigate pain and restore functions of injured human shoulders. The procedure involves the replacement of the affected shoulder ball and socket joint with the synthetically generated prosthesis. Long after the replacement of the affected part, the synthetic prosthesis maybe damaged or worn out, thus the reoperation process maybe repeated or revised. To guarantee a robust and seamless reoperation, detail of the model and manufacturer of the synthetic prosthesis is a paramount. There are circumstances where information regarding the prosthesis is not available. To determine the model and manufacturer, a thorough inspection and physical analogy of various manufacturers’ prosthesis are therefore required. The manual method consumes longer time and is liable to errors. With the evolution and continuous growth in the field of Machine Learning, prediction models can be used to learn the prosthesis and predict the model and manufacturer, thus reducing the time and error in the manual approach. In this paper, a model based on ensemble learning for identification of the manufacturer of synthetic prosthesis is proposed. Deep Convolution Neural Network (DCNN), High Resolution Network (HRNet), and Support Vector Machine (SVM) were combined to form the ensemble model. To ensure accurate prediction of the manufacturer, the components of the ensemble model are separately trained and then integrated using a novel weighted average ensemble technique. This strategy aids in identifying the prosthesis' manufacturer by assigning greater weight to the model that performs the best. The proposed model performance is compared with the performance of its individual components, where it outperforms the individual models in terms of accuracy, recall, precision and f measures.