Deep CNN-Based Ensemble CADx Model for Musculoskeletal Abnormality Detection from Radiographs

Abstract

Musculoskeletal Disorders (MSDs) are excoriations and afflictions that assail body movement of human. In present days diagnosis of musculoskeletal conditions are dependent on radiographs. Sometimes doctors or radiologist can make an error that can mislead the diagnosis of abnormalities. So, we have been motivated to develop a novel Computer-Aided Diagnosis(CADx) system based on Deep Convolutional Neural Network (Deep CNN) that will help the doctors to identify musculoskeletal abnormalities through radiographs. We have used VGG-19, ResNet architecture to build a model for four types of study (Elbow, Wrist, Finger, and Humerus). 5-fold cross-validation method is also applied to evaluate our models. Then we have applied ensemble techniques to improve the model’s performance. Finally, based on some performance evaluating metrics the best one is selected for each of the study types and in aggregate. Our proposed technique tested on a benchmark radiographic dataset named ‘MURA’, and the final result is compared to other prominent techniques. For Elbow, Finger, Humerus, Wrist study, model performance was consecutively 86.45%, 82.13%, 87.15%, and 87.86%. Experimental consequences show that our proposed method is a condign strategy to resolve musculoskeletal abnormalities detection.