Accurate segmentation of COVID-19 infected regions in lung CT scans with deep learning

The widespread impact of coronavirus disease 2019 (COVID-19) has led to a severe health crisis and loss of life affecting billions of people. Detecting COVID-19 early on and distinguishing it from other illnesses is a major challenge in the pandemic. Computed tomography scans are vital for diagnosis, but they are difficult and slow for radiologists to interpret. So, deep learning technology is being used to speed up disease diagnosis, especially for COVID-19. This research employs a multi-scale feature extraction module that captures features at different scales to detect small and large structures in infection areas. The channel attention module enhances feature representation by focusing on informative channels. A bidirectional pyramid module is utilized to acquire fine-grained details and reduce spatial dimensions. The global dependencies and long-range contextual relationships among the images are captured through global and edge paths, resulting in clear and accurate images. Finally, the outputs from both modules are merged in the feature fusion Network to provide an accurate segmentation mask for COVID-19-related areas. The testing is conducted using the COVID-19 lung scan image dataset. Extensive trials are performed using standard metrics, and the output is compared with recently proposed approaches. The results demonstrate that this method achieves 98.6% accuracy, 96.54% precision, 97.52% sensitivity, 98.55% specificity, 98.59% kappa score, and an execution time of 3.2 s. These results suggest that this method performs better than other comparative methods and accurately identifies regions in the lungs affected by COVID-19.