Abstract
Deep vein thrombosis (DVT) is a potentially life-threatening condition where a blood clot forms in deep veins, typically in the calf muscle region of our lower extremities. It may cause leg pain or swelling, and it might break down, resulting in pulmonary embolism (PE). When DVT and PE occur concurrently, it is commonly referred to as venous thromboembolism (VTE). This research develops an innovative methodology for DVT segmentation in medical imaging. Using a novel approach, we provide an updated U-Net architecture that includes residual and inception blocks to improve the model's capacity to precisely identify DVT areas. In addition, we investigate the use of Convolutional Neural Networks (CNN), for semantic segmentation in order to obtain an in-depth segmentation of DVT. In comparison to conventional approaches, our results illustrate that the proposed architecture is effective in recognizing DVT edges with better accuracy and resilience. Our model is trained using binary cross-entropy loss and the Adam optimizer and evaluated using performance metrics like accuracy, loss, and sensitivity. Our research enhances the area of medical image processing by presenting a potential path for automated and robust techniques in DVT segmentation.
