Enhancing Model Performance in Detecting Abnormalities on Human Musculoskeletal System through Optimization Techniques using Deep Learning
Musculoskeletal Radio-graph studies have gained much attention over the recent years in the field of Medical Science and with the advancement of Deep Learning, it has gained high prominence in terms of Classification of Disease or Abnormalities in the Human Musculoskeletal Systems. In this paper, we have taken a MURA Dataset (MURA stands for Musculoskeletal Radio-graphs) and based on the previous studies by the Radiologists, we have classified the Radiograph Images as either Normal or Abnormal for Seven Categories of Studies. Those Seven categories are Elbow, Finger, Forearm, Hand, Humerus, Shoulder and Wrist. Also we have generated the AUROC curve for the results we have obtained after training the Model with the classifier. In our work, we have also used different combination of Network Architecture with the use of ADAM optimizer to compare & analyses the best one that can optimize the Model to provide the best possible results in case of MURA Dataset.
Keywords: Musculoskeletal Radio-graphs, Abnormality, Deep Learning, Network Architecture, Classifications.