Automated Detection of Shoulder Arthroplasty in X-rays Using Machine Learning

Abstract

Demand for shoulder arthroplasty is rising at a faster rate than hip and knee arthroplasty, driven by an increasingly aging yet active population. Joint registries are playing an increasingly critical role in tracking the long-term success of shoulder arthroplasty, identifying failure mechanisms, and shaping clinical best practices but current classification procedures are often performed by non-medically trained encoders leading to error. This study examines the use of machine learning in techniques to classify four broad categories of shoulder arthroplasty technique from postoperative x-rays. Data from the Scottish Arthroplasty Project, was used to create a balanced dataset of 1000 samples. A 10-fold cross validation was used for the training of 4 neural network models commonly used for classification of x-ray data. InceptionV3 model achieved the highest overall performance with an accuracy of 93.85% after cross validation, while EfficientNet demonstrated the highest individual classifier accuracy of 99% suggesting the potential to increase accuracy further in future studies.Clinical Relevance- This research highlights the potential of machine learning to enhance the accuracy of joint registry data encoding. Facilitating evidence-based improvements in implant design and surgical approaches through the use of more accurate data on implant survival, and revision rates.

Ähnliche Arbeiten