Seeker optimization with mask RCNN based efficient model for Covid-19 detection and severity analysis using CT images

Accurate and timely detection of Covid-19 from medical imaging plays a critical role in patient management and public health response. This research presents a pioneering approach for Covid-19 segmentation and severity analysis from CT images, leveraging a novel Seeker optimization with a mask region convolutional neural network (Mask RCNN) based efficient model. The methodology integrates advanced optimization techniques with state-of-the-art deep learning architectures to achieve precise segmentation of Covid-19 lesions and comprehensive severity assessment based on the spread area of the disease. The proposed framework begins with an adaptive filtering pre-processing step applied to the CT images, aimed at reducing noise and enhancing relevant features crucial for subsequent analysis. This pre-processing step employs adaptive filtering techniques to adjust filter parameters based on local image characteristics, thereby improving image quality and detection accuracy. Subsequently, Seeker Optimization employs to optimize the hyperparameters of the Mask RCNN model, including network architecture, learning rates, and regularization parameters. Seeker optimization algorithm (SOA) dynamically adapts to the complex parameter space, efficiently guiding the Optimized mask region convolutional neural network (OMRCNN) model towards optimal configurations for accurate Covid-19 detection and segmentation. The OMRCNN model is then utilized for Covid-19 detection and segmentation, focusing on precise localization and delineation of infected regions within CT scans. Leveraging its ability to generate high-quality segmentation masks, the proposed approach facilitates the quantitative assessment of disease severity by analyzing the spread area of Covid-19 lesions. Severity analysis is performed by quantifying the extent of lung involvement, providing valuable insights into disease progression and informing clinical decision-making. To evaluate the efficacy of the proposed OMRCNN methodology, extensive experiments were conducted on a diverse dataset of CT images from Covid-19 positive cases. The dataset included 2481 CT images, divided into 1252 coronavirus-infected images and 1229 non-coronavirus images. Among these, 1985 images were used for training and the remaining for testing. The model achieved an accuracy of 96.73% in detecting Covid-19 disease. Additionally, it showed significant improvement in segmentation performance, with metrics such as precision, recall, sensitivity, specificity, and F1-Score outperforming conventional approaches. By accurately delineating Covid-19 lesions and quantifying disease severity, the framework offers a valuable tool for clinicians in diagnosing and managing Covid-19 patients effectively.