<p>Pulmonary Embolism (PE) is a dangerous illness brought on by a sudden obstruction in the pulmonary artery, which can be fatal without early detection. During segmentation of pulmonary embolism through deep-learning and machine learning techniques, loss function plays a vital role in appropriate segmentation. This study focuses on the selection and application of various loss functions tailored for PE segmentation. In this work, the publicly available PE challenge dataset, which comprises chest computed tomography (CT) scans with pulmonary embolism, is used. UNet based segmentation models are evaluated with different loss functions. The results demonstrate that both Dice Loss and Structural Similarity Index measure (SSIM) Loss significantly enhance the segmentation outcomes, with the UNET(U shaped convolution network) model outperforming the others. The UNET model achieves a sensitivity of 0.8673, a specificity of 0.9989, a Dice coefficient of 0.9259, a Jaccard similarity index of 0.9120, and an accuracy of 0.9889, while also requiring less computational time and resources. These results highlight the importance of choosing the appropriate loss function to increase the efficiency and precision of PE detection segmentation. This work highlights the critical role of loss functions in optimizing UNet-based architectures for PE segmentation, providing valuable insights for the development of more effective diagnostic tools.</p>