การจำแนกเนื้องอกบริเวณอุ้งเชิงกรานจากภาพถ่าย MRI โดยใช้การเรียนรู้เชิงลึกด้วยสถาปัตยกรรม U-Net ร่วมกับขบวนการโมโฟโลจี

Abstract

Pelvic tumors are a condition with complex physical characteristics, making cancer diagnosis challenging. Magnetic Resonance Imaging (MRI) is a technique used for tumor segmentation to assist in treatment planning. However, there are still limitations regarding the number of images and the expertise of medical personnel. Due to the low incidence of the disease, there is a shortage of trained professionals, which leads to delayed diagnosis and treatment for patients. This research presents a method for segmentation of pelvic tumors from MRI images using deep learning with a U-Net architecture combined with Morphological Operations to facilitate medical personnel in diagnosis and treatment planning to suit the patients best. The approach also aims to improve the accuracy of pelvic tumor segmentation. The algorithm tests demonstrated that the deep learning method using U-Net architecture combined with Morphological Operations achieved an evaluation performance with Dice Similarity Coefficient (DSC) and Intersection Over Union (IoU) of 93.30% and 87.44%, respectively. This experiment yielded better results than the study by Y. Qu et al. [6], which proposed cancer segmentation for surgical planning from MRI and CT images using a multi-view fusion network. The performance achieved a DSC score of 85.51%.