Advanced image segmentation and severity prediction for COVID-19 using nnU-Net and optimized FRCNN-GJS algorithm

The global spread of Coronavirus Disease 2019 (COVID-19) has led to infections in millions of individuals with infected patients facing the risk of death as symptoms progress. Early identification of severely sick patients is crucial for timely intervention. This study presents a technique for image segmentation and disease prediction based on severity levels. The process starts with preprocessing steps, including denoising, grayscale binarization, Gaussian blur, and image flipping to enhance image quality. The No new U-Net (nnU-Net) model is employed for accurate semantic image segmentation, to identify lung regions. Subsequently, a novel Faster Region-based Convolutional Neural Network-Golden Jackal Search (FRCNN-GJS) algorithm is introduced for predicting disease severity based on the lung infection percentage. This approach classifies severity levels into five grades: healthy, mild, moderate, severe, and critical, thus increasing diagnostic capabilities for medical professionals. The integration of the GJS algorithm optimizes hyperparameters, which improves model performance and accuracy. The experimental results indicate the accuracy of 98.84%, Dice Similarity Index (DSI) of 0.916, and Mean Absolute Error (MAE) of 0.324 which outperforms existing methods. The model shows the lowest prediction time and efficient resource utilization, aiding COVID-19 patients and minimizing the strain on healthcare systems.