From CNN to Transformer: A Review of Medical Image Segmentation Models

Wenjian Yao,Jiajun Bai,Wei Liao,Yuheng Chen,Mengjuan Liu,Yao Xie

2023-08-10

Abstract:Medical image segmentation is an important step in medical image analysis, especially as a crucial prerequisite for efficient disease diagnosis and treatment. The use of deep learning for image segmentation has become a prevalent trend. The widely adopted approach currently is U-Net and its variants. Additionally, with the remarkable success of pre-trained models in natural language processing tasks, transformer-based models like TransUNet have achieved desirable performance on multiple medical image segmentation datasets. In this paper, we conduct a survey of the most representative four medical image segmentation models in recent years. We theoretically analyze the characteristics of these models and quantitatively evaluate their performance on two benchmark datasets (i.e., Tuberculosis Chest X-rays and ovarian tumors). Finally, we discuss the main challenges and future trends in medical image segmentation. Our work can assist researchers in the related field to quickly establish medical segmentation models tailored to specific regions.

Image and Video Processing,Computer Vision and Pattern Recognition,Machine Learning

What problem does this paper attempt to address?

The paper mainly addresses issues in the field of medical image segmentation, particularly focusing on improving the accuracy and efficiency of image segmentation using deep learning techniques. Specifically, the paper focuses on the following aspects:

1. **Importance of Medical Image Segmentation**: Medical image segmentation is a crucial step in medical image analysis, playing a vital role in the efficient diagnosis and treatment of diseases.

2. **Limitations of Existing Methods**: Traditional image segmentation methods rely on manual feature extraction or handcrafted algorithms based on image processing techniques. These methods have limitations in terms of efficiency and accuracy when dealing with large volumes of complex medical images.

3. **Application of Deep Learning in Image Segmentation**: In recent years, deep learning-based image segmentation methods have become the mainstream trend. Convolutional Neural Networks (CNN) and their variants, such as U-Net, are widely adopted. However, CNNs have limited capability in handling long-range dependencies.

4. **Application of Transformer Models**: With the success of pre-trained models in natural language processing tasks, Transformer-based models have also been introduced into medical image segmentation tasks, showing excellent performance.

5. **Main Contributions of the Paper**: This paper reviews four of the most representative medical image segmentation models in recent years, including U-Net, UNet++, TransUNet, and Swin-Unet. The paper not only theoretically analyzes the characteristics of these models but also quantitatively evaluates their performance on two benchmark datasets (chest X-ray images of tuberculosis and ovarian tumor images). Additionally, the paper discusses the main challenges and future development trends in the field of medical image segmentation and provides all experimental codes and detailed model configuration parameters to help researchers in related fields quickly understand and apply these models.

In summary, this paper aims to promote the advancement and development of medical image segmentation technology through a comprehensive evaluation of current representative medical image segmentation models, particularly in improving segmentation accuracy and model generalization capabilities using Transformer structures.