Assessing the Performance of U-Net in Three Dimensional Medical Image Segmentation

EasyChair Preprint 11489

Abstract

3D medical data plays a vital role in the field of healthcare, particularly in disease diagnosis and surgical planning, as it allows for precise identification, segmentation, and visualization of lesions and organs. However, organ segmentation in medical images is a challenging task due to the diverse features of organs, such as variations in size, shape, and location. To address this challenge, deep learning techniques have been widely utilized for medical image segmentation.One of the most prominent and successful algorithms used for this purpose is U-Net, which was proposed in 2015demonstrated significant achievements in various medical segmentation tasks and has gained popularity among researchers and scientists. Its effectiveness and versatility have led to its continuous development and improvement in recent years. With over 70,000 citations, U-Net has become a widely adopted algorithm for medical image segmentation.To evaluate the performance of U-Net and ensure more accurate and detailed segmentation results, it is commonly trained on 3D datasets containing deformations of organs, such as the spleen, obtained from the Medical Segmentation Decathlon. The evaluation of the algorithm's capabilities often relies on metrics like the Accuracy and f1-score , which is a widely accepted measure for assessing the quality of medical image segmentation.

Keyphrases: Segmentation, dimensional, image, medical, three

@booklet{EasyChair:11489,
  author    = {Fatma Khenaifer and Ilyes Benaissa and Athmane Zitouni and Zine-Eddine Baarir},
  title     = {Assessing the Performance of U-Net in Three Dimensional  Medical Image Segmentation},
  howpublished = {EasyChair Preprint 11489},
  year      = {EasyChair, 2023}}