Enhancing Pulmonary Embolism Segmentation Through Optimized SwinUnet with Resnet 152

Pulmonary embolism arises when blood clots migrate to the lungs. The fusion of computer vision and deep learning enhances diagnostic capabilities, supporting medical research goals of reducing mortality rates, optimizing healthcare costs, and improving treatment outcomes. Despite progress, challenges persist in accurately segmenting pulmonary embolism, with current methods prone to underfitting and overfitting issues. To address these challenges, this research introduces an innovative approach to enhance pulmonary embolism segmentation. By leveraging the Radiological Society of North America Society of Thoracic Radiology (RSNA STR) pulmonary embolism detection dataset, strategically employ Color Wiener Filtering (CWF) to enhance image quality during preprocessing. The SwinUNet model, with its Swin Transformers attention mechanism, generates detailed segmentation crucial for accuracy. The hybrid SwinUNet with ResNet 152 (SURN-152) combines features, especially beneficial for intricate details in pulmonary embolism. Kepler optimization fine-tunes SURN-152 parameters, significantly improving segmentation capabilities. Evaluation metrics, including precision of 99.33%, accuracy of 99.95%, recall of 99.66% and specificity of 99.12%, ensure robust validation using Python, providing insights into segmentation accuracy.