Diagnosing Musculoskeletal Disorders from Shoulder Radiographs Using Deep Learning Models

Abstract

In recent years, exploring how Artificial Intelligence, especially deep learning algorithms, may be used to analyse healthcare data has become crucial. Deep learning can be employed for assisting the doctor and supplementing the process of diagnosing diseases. Early diagnosing of the problems of the shoulder, a vital body part will be useful to manage or solve the shoulder pain with efficacy. Our study has utilized deep learning to diagnose abnormalities from radiographs of the upper extremity, specifically the shoulder. The abnormality in the shoulder has been detected from the radiographs of the shoulder using four pre-trained Convolutional neural networks (CNN) models such as VGG-16, DenseNet-201, ResNet-50, and EfficientNet-B1. We have also calculated the accuracy, precision, recall, specificity, sensitivity, Flscore, and area under the receiver operating characteristic (AUROC). Shoulder radiographs from the MURA dataset were used to test the four CNN models. From the experiments conducted, both ResNet-50 and VGG-16 had the maximum accuracy of 81%; ResNet50 had the best AUC score of 0.74 and it outperformed the other three models.

Ähnliche Arbeiten