The chest X-ray imaging (CXR) is a key diagnostic instrument in COVID-19 diagnosis, wherein more than 600,000 tests are performed worldwide annually and the misdiagnosis rate is estimated to be 15-20 percent, largely contributed by human error. Conventional manual reading of CXR images is time-consuming, labor-intensive, and heavily reliant on the skill of the radiologist, typically resulting in a series of uneven and sluggish diagnostic outcomes. To overcome these limitations, the current research introduces an innovative state-of-the-art CXR segmentation model based on rigorous preprocessing techniques in combination with the optimisation of deep-learning algorithms to obtain precise lung parenchyma and pathological lesion outlines. Block-matching 3D filtering (BM3D) was applied to suppress noise without loss of anatomical details following curation of the COVID-19 CXR Dataset. The Optimization U-Net (OU-Net) architecture, which served as the backbone of the proposed approach, was carefully designed with adaptive encoder-decoder paths and strengthened skip connections to better subdivide real lung regions and manifestations of diseases. Additionally, the training schedule utilizes Modified Grey Wolf Optimization (MGWO) for the optimization of network parameters, and this accelerates convergence and enhances segmentation accuracy. Empirical results confirm that the OU-Net with MGWO is superior to conventional and standard deep-learning models, as the suggested approach enhances accuracy by 4.58%, sensitivity by 5.22%, specificity by 4.60%, precision by 4.85%, recall by 1.78%, F1-score by 5.07%, Jaccard index by 5.23%, and Dice score by 5.31%.

Block-matching and 3D Filtering COVID-19 Chest X-Ray Grey Wolf Optimization Optimal U-Net Segmentation