Using Convolutional Neural Networks to Analyze X-Ray Radiographs for Multi-Label Classifications of Thoracic Diseases

Abstract

Currently, it takes approximately 6 to 8 weeks from the initial doctor's examination to diagnose lung disease. This could potentially lead to the patient's condition worsening, the disease becoming unmanageable, or may lead to the patient's death. In order to aid doctors in the accurate and more timely diagnosis of their patients, we propose the use of convolutional neural networks for computer-aided diagnosis. Our application uses image recognition to identify the traits of various diseases in radiographs to successfully diagnose a patient. This is done through training a CNN with a dataset of 112,120 images of lung diseases. The model was tested with a resulting validation accuracy of 93 percent. The application will benefit patients suffering from these illnesses as it is time-efficient, cost-effective, and more accurate than manual diagnosis.