Abstract:
Computer aided diagnosis plays a key role in improvement and better understanding of the medical imagery in order to address problems faced in the field of medical diagnosis. Medical imagery is used to view and analyze different types of scans performed on the body, these scans could be CAT scans, X-ray scans, MRI scans etc. Medical errors such as misdiagnosis is one of the leading causes of injuries and deaths worldwide. A radiologist has a subjective view when it comes to diagnosis from medical images and sometimes this results in a false positive/negative due to various reasons such as his workload, level of fatigue etc. Advancements in medical image processing and machine learning has made it likelier for a system to exist that assists in diagnosis of scan images, providing a second opinion on the findings a radiologist has deduced. Deep Learning, a subset of machine learning concerned with algorithms inspired by structure of the brain called artificial neural networks (ANN) will be utilized and in a sense constructed to provide us the best way for automatic detection and classification of bone abnormalities for 6 different types of bones namely wrist, shoulder, finger, elbow, hand and humerus from X-ray scan images. Each bone dataset we have consists of multiple patients with their studies associated. The main purpose of the project is to research and build an instant bone abnormalities classification system using X-ray scan images based on a deep learning model. We aim to research and apply different state-of-the-art CNN architectures based on image classification and modify them to build a model that best suits our application. The system helps differentiate normal and abnormal X-Rays which is expected to increase the efficiency in terms of radiologist handling cases. System will have X-ray images go through a pre-screening process in order to filter out cases that are less serious. We aim for a web-based system where an X-ray image could be uploaded for instant diagnosis. This assists doctors in remote areas where access to skilled radiologists is limited and process of diagnosis is time costly.
