Lymphoma Lesion Segmentation in Neck Axial and Coronal Section Ct Images with U-Net

Abstract

Lymphoma is a type of cancer classified under blood cancer, emerging because of cell degradation. It is visualized using computed tomography (CT) and positron emission tomography (PET). Segmentation refers to the process of separating the desired object from the background and other elements in an image. In this study, CT images obtained from lymphoma patients were segmented using U-Net, a deep learning-based segmentation model. The test results achieved were a 0.83 Dice Similarity Coefficient (DSC), 0.73 Jaccard Index (JI), an average Hausdorff Distance (HD) of 15.81 mm for neck axial CT images, and a 0.75 DSC, 0.64 JI, and an average HD of 30.18 mm for neck coronal CT images. The detailed statistical and visual analyses demonstrated that the lymphoma lesion segmentation was successfully performed on neck coronal and axial CT images using U-Net.