Chest Disease Classification Using Convolutional Neural Networks

Abstract

Chest diseases pose major health risks to people globally. Early diagnosis of these conditions enables early treatment, which can prevent death. The healthcare system benefits greatly from the use of Convolutional Neural Networks (CNN), particularly when it comes to early disease prediction. X- rays serve as one of the key factors that accurately classify disorders of the chest. The prediction of chest diseases such Atelectasis, Cardiomegaly, Lung Consolidation, Edema, Pleural Thickening and Pneumothorax from X-ray images is the objective of this project. The CNN Model is used to analyze the disease classification and the results are explained in terms of accuracy. Preprocessing with images can enhance the model’s accuracy. For that, we used some image preprocessing techniques which include Histogram Equalization, Bilateral Filter, Gaussian Blur and Contrast Limited Adaptive Histogram Equalization. These techniques were used to remove the unwanted noise from the X ray images and improve luminance of the images which leads to produce more accurate decisions. The dataset consists of 1 csv file and an X-ray image folder that contains six classes of disease and 1,120 X-rays. Convolutional neural networks (CNNs) are described in the research as a tool for diagnosing disorders of the chest. The architecture of CNN is presented, as well as its guiding principles. Among those preprocessing techniques, Contrast Limited Adaptive Histogram Equalization technique gave more accuracy which is nearly 91.2 %. Results that compare accuracy and network training time are shown.