Determining whether a radiographic study is normal or abnormal is a critical radiological task: a study interpreted as normal rules out disease and can eliminate the need for patients to undergo further diagnostic procedures or interventions. The musculoskeletal abnormality detection task is particularly critical as more than 1.7 billion people are affected by musculoskeletal conditions worldwide (BMU, 2017). These conditions are the most common cause of severe, long-term pain and disability (Woolf & Pfleger, 2003), with 30 million emergency department visits annually and increasing.
We have used the MURA dataset, contains 9,045 normal and 5,818 abnormal musculoskeletal radiographic studies of the upper extremity including the shoulder, humerus, elbow, forearm, wrist, hand, and finger. MURA is one of the largest public radiographic image datasets.
More information about the the dataset can be found here as well as in this paper
We have implemented deep learning based models for abnormality detection. (Code coming soon)