Classification of Chronic Obstructive Pulmonary Diseases from Chest X-Ray Images Using Deep Learning

Abstract

To reduce the risk of Chronic Obstructive Pulmonary Disease and we have proposed the applications of digital image processing techniques. To accomplish our study, we have adopted a design science methodology and followed its scientific procedures starting from collecting the required data set to test the developed model. We have collected about 2248 images having 350 Images or more for each class. We have applied different image preprocessing tasks to enhance the image. And augmentation is applied to increase the number of images to a total of 2248 chest X-ray Images. Therefore, to overcome that problem, we applied zooming, rotation, and flipping at a different angle as augmentation techniques. Then Features are extracted from gray-level images using a CNN feature extraction and a classification model is built using 5 Different Pre-trained models called InceptionV3, VGG16, EffeceintNetB0, and Resnet50 including our own CNN model. The convolutional neural network architecture with the sequential model is implemented with many layers such as convolutional, activation, max-pooling to extract important features from the Chronic Obstructive Pulmonary diseases x-ray image. A total of 2248 COPD Chest X-ray datasets were collected from St. Paul Millennium Medical College Hospital Black Lion Specialized Hospital, Betele Specialized hospital, ReftyVally University Collage Specialized Hospital, MSF Holland Medical Center Gambella Branch, and Jimma University Medical Center. An adequate set of a report for labeling was not available and requires tremendous effort and time. We have used 80/20 by splitting the data into 80% for training and 20% for tests. Transfer learning and data augmentation techniques were applied. The proposed CNN classification model achieved an average accuracy of 81.1%. While the InceptionV3 with its filtering mechanism has achieved a better classification performance with an accuracy of 90.1% and it was reported as Highly Accuracy we obtained through the investigation. The study could contribute to the medical profession by providing a system that supports experts to estimate chronological age