Pulmonary Disease Identification and Classification using a Deep Learning Approach

Abstract

Deep Learning (DL) models have shown strong results in detecting diseases from medical images. In this paper, we explored the challenge of classifying pulmonary diseases (PD) using chest X-rays. The research focuses on three common respiratory conditions: pneumonia, pulmonary tuberculosis, and pleural effusion. We proposed a new framework for detecting and classifying PD from chest X-ray (CXR) images. The process includes noise reduction, image enhancement, data augmentation, segmentation, feature extraction, and classification. To remove noise, we used a Gaussian filter, and for improving image quality, we applied an advanced histogram equalization method. The Region of Interest (ROI) of the lungs was extracted using Otsu’s threshold segmentation technique. For feature extraction, we used the Gabor filter to obtain texture details from the images. A Deep Convolutional Neural Network (DCNN) was then used for classification. The system classifies images into four categories: normal, pneumonia, pulmonary tuberculosis, and pleural effusion using a four-way SoftMax classifier. We tested four DCNN models: VGG16, VGG19, ResNet50V2, and DenseNet201. Among these, DenseNet201 performed best, achieving a training accuracy of 97.80% and a testing accuracy of 95.73%. Compared with other advanced models, DenseNet201 showed higher accuracy and better capability in detecting and classifying pulmonary diseases.

Keywords: Feature Learning, Gabor Filter, Pulmonary Disease, Segmentation, X-Ray, Respiratory Conditions