This research paper explores the application of deep learning techniques for the automated classification and segmentation of COVID-19, Normal, and Viral pneumonia cases using chest X-ray images. The dataset comprises 510 grayscale chest X-ray samples collected from publicly available COVID-19 repositories, equally distributed across three categories. The primary objectives of this study include identifying COVID-19 infection patterns, enhancing medical image classification performance, and providing a visual interpretation of model outputs for clinical utility. The methodology integrates image preprocessing and normalization followed by unsupervised k-means clustering to observe data distribution. A U-Net model is employed for pixel-level segmentation to highlight infection regions, while hybrid CNN and LSTM architecture is developed for image-level classification. The classification model achieved a test accuracy of 74.5%, with a precision of 97% for COVID-19 class and strong macro average scores, reflecting balanced performance across all classes. Results are visually represented using segmentation overlays, a confusion matrix, and bar plots for class distributions. This integrative approach supports early detection and decision-making in clinical settings, combining segmentation clarity with reliable classification metrics.
Arunadevi et al. (Mon,) studied this question.