Abnormality Detection in Musculoskeletal Radiographs

Abstract

A radiographic study is a typical procedure utilized to find different kinds of abnormalities, where the identification of Musculoskeletal Abnormalities from the norm stands out as a crucial task. in this paper we suggests some deep learning methods to identify such Musculoskeletal Abnormalities using the MURA dataset which is one of the largest collections of upper extremity radiographs. We make use of DenseNet and VGG deep learning models to try to draw some unprecedented findings and results. In addition, we try to visualize all our results and findings using comparative graph study. We compare our model with models proposed during the Stanford ML Group MURA Competition using Cohen Kappa Statistic. The obtained results show that deep convolutional neural networks can achieve results which are very close and even better compared to current state-of-the-art models. We achieve comparable model performance to state-of-the-art performances in three of seven study types.