Automated Make and Model Identification of Reverse Shoulder Implants Using Deep Learning Methodology

Identification of an Orthopaedic Implant before a revision surgery is very important. Failure to identify an implant causes surgical planning delays, inability to plan for the correct equipment requirements, and can result in poorer patient outcomes. This paper proposes a framework to identify, make and model of two different reverse shoulder implants from X-ray images using Deep Learning Techniques. Both Anterior Posterior and Lateral views of X-rays were used in the study and a comparison was made to identify which view enables better results in identification. Various pre-trained deep learning models such as VGG16, VGG19 and InceptionV3 were used for classification of implants. The proposed methodology identifies both the make and model of the implant with an accuracy of 95% using both Anterior Posterior and Lateral Views and an accuracy of 86.67% using only the Anterior view.