Advancements in Automated Segmentation Techniques for Endometrial Cancer: A Comprehensive Review

Endometrial cancer is the most common gynecological cancer that requires an accurate detection and diagnosis for effective treatment. Automated segmentation of medical images, MRI(Magnetic Resonance Imaging), PET (Positron Emission Tomography), and CT (Computed Tomography) scans, raises diagnostic accuracy, reduces human error, and assures consistent identification of tumor margins, hence, it provides early detection and personalized treatment strategies. The study focuses on several automated segmentation techniques, with particular attention on the effectiveness of U-Net Architecture for MRI images with the achievement of high Dice coefficients (0.959 for uterus and 0.911 for tumor segmentation).PET/CT images, on the other side, showed the highest Dice ever recorded was 0.739 with the PET images and 0.510 with the CT images, substantiating the difficulties of segmentation in multi-modality datasets. The low performance of automated Segmentation in PET and CT Scans, as reported in the literature, led us to review various hybrid cnn-transformer models that have demonstrated effectiveness in other tumor segmentation tasks. This study delves into the potential of imaging PET and CT and introduces promising future directions to enhance performance metrics of automated segmentation of PET and CT images.