Classification Between Normal and Abnormal X-Rays

Abstract

The prevalence of x-ray examinations for diagnosing various medical conditions worldwide, with around 400 million which are conducted annually according to the World Health Organization, underline the significance of this imaging modality. Around 30 million x-rays alone in Pakistan are performed annually. This huge number of x-rays which are generated, poses a crucial challenge of timely and accurately reporting due to shortage of radiologists. This research aims at pressing need of an automated system which is capable of detecting between normal and abnormal x-rays in a manner which allows radiologists to divert their attention towards abnormal x-rays. The huge workload in larger healthcare organizations, which is coupled with a lack of experienced medical radiologists in underdeveloped countries, undermines the importance and need for an accurate and efficient classification system. This research utilizes deep learning, which is a subfield of machine learning, has demonstrated outstanding success in various domains, including medical imaging analysis. Especially, convolutional neural networks (CNNs) which is also known for their productiveness and effectiveness in computer vision related tasks. This research aims of utilizing state-of-the-art image classification algorithm, ResNet50, which is used to train the model, and aims to achieve higher accuracy of 85% on the pretrained dataset. Deep learning algorithms is known to play a vital role in accurately identifying and efficiently classifying x-rays images as normal or abnormal. Automating the entire classification, will aim to alleviate burden of radiologists and will enable them to focus more on the abnormality found in x-rays. This automation will extend to many other medical conditions including breast cancer, orthopedic conditions, lung infections, dental issues, which will make it extensive advancement in field of medical imaging. In this research, we put into practice a binary classification approach which analyzes chest x-rays and categorizing them into two different distinct classes: normal x-ray images and x-ray images that are abnormal. This classification model serves as a fundamental procedure for differentiating between X-ray scans that are in healthy conditions and those that show many abnormalities or deformity within the specified region. This research is supported by healthcare organizational collaboration of SLOSH AI Solutions.