A Deep Learning Paradigm for Computer Aided Diagnosis of Emphysema from Lung HRCT Images

Emphysema is associated with lung tissue loss and is one of the primary reasons for COPD (Chronic Obstructive Pulmonary Disease). Thus, to detect the stage of COPD and proliferation of the emphysematous lesions, it is necessary to distinguish between emphysematous and healthy lung tissues. This study discusses a novel approach to determining the lungs’ emphysema through chest C.T. image analysis. For this, an intensity threshold of −910 HU is taken to detect the emphysematous tissues from the affected lungs highlighted on the C.T. images and further segregated into normal and emphysema affected lung C.T. images to bring automation in classification and emphysema detection using Deep Learning algorithms. C.T. images of 115 subjects (COPD and non-COPD smokers and non-smokers) were processed and used for classification in this study to screen out emphysema lung conditions. Because of such a small database building, a CNN from scratch yielded an accuracy of 54% in classification, which is why transfer learning was used to enhance the ability of emphysema detection. Furthermore, pre-trained network VGG16 used to classify the C.T. images outperformed the basic CNN with a whooping accuracy of 88% for determining emphysema.