Pelvic Tumor Segmentation in Magnetic Resonance Images By U-Net

Abstract

Magnetic Resonance Imaging (MRI) is widely used for diagnosing pelvic tumors. Tumor segmentation is an essential step in surgical planning; however, it typically requires specialized experts to manually segment tumors in MRI scans. This research presents a method for segmenting pelvic tumors from MRI images using deep learning technique. A U-Net framework and a U-Net with Batch Normalization architecture for pelvic tumor segmentation are used. To test the performance of the proposed method, we use 24 T2-weighted (T2W) MRI images. The Dice Similarity Coefficient (DSC) and Intersection Over Union (IoU) of the U-Net are 71.99% and 56.24%, respectively, while the DSC and IoU of the U-Net with batch normalization layer are 88.68% and 79.66%, respectively. The U-Net with batch normalization layer shows satisfactory results.