A Deep Learning Framework for Detection and Classification of Implant Manufacturer using X-Ray Radiographs

Now-a-days, artificial prosthesis is widely used to mitigate pain in damaged shoulders and restore their movement ability. The process involves a complex surgery that attempts to fix an artificial prosthesis into a dead shoulder as a replacement for the ball and socket joints of the shoulder. Long after the surgical process, the need for revision or reoperation may arise due to some problems with the prosthesis. Identification of prosthesis manufacturer is a paramount step in the reoperation exercise. Traditional approach compares the prosthesis under consideration with prosthesis from a vast number of manufacturers. This approach is cost-efficient and requires no extra training for the physician to identify the prosthesis manufacturer. However, the method is time inefficient and is prone to mistakes. Systems based on machine learning have the potential to reduce human errors and expedite the revision process. This paper proposes a shallow 2D convolution neural network (CNN) for the classification of shoulder prosthesis To speed-up the learning process and improve the performance of the deep learning model for implant classification, this paper employed three different techniques. Firstly, a generative adversarial network (GAN) is applied to the dataset to augment the classes with fewer samples to ensure the data imbalance problem is eliminated. Secondly, the highly discriminating features are extracted using principal component analysis (PCA) and used to train the proposed model. Lastly, the model hyper-parameters are optimised to ensure optimal model performance. The model trained with extracted features with a variance of 0.99 achieved the best accuracy of 99.8%.