Advancing COPD Diagnosis through Deep Learning, GANs, and Chest X-Ray Analysis for Precise Detection and Severity

The rate of mortality within those who have COPD (chronic obstructive pulmonary disease) has risen recently, and itis predicted that it willstayon the risein the years to come. Due to the extensive laboratory testing requiredforconfirmation, standardapproachestakeavery long time to diagnose these medicalconditions. Fortunately, with the development of sophisticated approaches and consideration of the ability to develop powerful techniques forpredictingothercriticaldiseases, itisanticipatedthatthis will helptopreciselyandpromptlydetectchronicdiseases.To aid in the diagnosis of COPD, this project aims to create a clinicaldecisionsupportsystembasedondeeplearning(DL) algorithms. Chest X-rays from 100 patients for COPD diagnosis were selected for examination. Chest X-rays are frequentlyutilisedtofindanomaliesthataidinthedetection of COPD. Many X-ray images are needed to traina classification model; however, radiographic Chest X-rays are still limited in availability. This may hinder the effectivenessoftechniquesforCOPDidentificationthatrely ondeeplearning(DL).Inthis study, thegeneratedChestX- rays using a Generative Adversarial Network (GAN) and Data Augmentation is used to get around these restrictions. Each X-ray that is produced falls into one of the two categories: COPD or normal. Now, the GAN-generated syntheticimageswerecombinedwiththerealimagesandfed into the following CNN architectures – VGG16, ResNet, DenseNet, Inception. Amongthesemodels, theoutputfrom the best-performing model is used to identifytheCOPD- affectedpatients. Further, their FEV1 factor range is considered to classify them depending on the seriousness of COPD – Mild, Moderate, Severe and Very Severe.