Artificial Intelligence based identification of Total Knee Arthroplasty Implants

Abstract

The identification of the make and model of the primary knee implant is an essential step for planning a revision surgery. Currently, the surgeons email the radiographs of the implant to the medical representatives of the manufacturing companies to get this information. This manual process is prone to errors and involves a considerable amount of resources and surgeon time that could be otherwise spent on more useful tasks. This study proposes an artificial intelligence based solution for the automatic identification of 6 makes of orthopedic knee implants. Deep Convolutional Neural Networks were trained on a dataset comprising 878 images of radiographs of orthopedic knee implants including both anterior posterior and lateral views. The results of the experiments showed a validation accuracy of 96.66%. Furthermore, class activation maps generated on the images when passed through the deep learning algorithm provided visual conformance to the region of interests that a surgeon would consider for identification of the make of the implant. The outcomes of this study demonstrates the effectiveness of the usage of Deep Convolutional Neural Networks for assisting orthopedic surgeons in pre-operative planning of revision surgery by accurate automated identification of make and model of orthopedic knee implant.

Ähnliche Arbeiten