Detecting Different Thoracic Disease Using CNN-Model

Use of CRX or chest X-ray images is the only way to perform the diagnosis of certain diseases, which can be difficult at times, thus it is crucial to interpret the displayed images with hundred percent accuracy. As a result of this issue, the usage of various algorithms and techniques to diagnose diseases automatically has increased. This research paper focuses on how deep learning, specifically the Convolutional Neural Network (CNN), can be used to diagnose thoracic disorders. We proposed customized CNN models with fine tuning, and applied the entire DenseNet architecture in CNN, including DenseNet-121, DenseNet-169, and DenseNet-201. The dataset included 14 thoracic disorders with no findings, and we chose six of them (Pneumonia, Emphysema, Pneumothorax, Fibrosis, Infiltrate and Pleural Thickening). With the DenseNet-121, DenseNet-169, and DenseNet-201 design, we were able to achieve accuracy of 0.839, 0.835 and 0.830 respectively.