Deep Learning-based Semantic Segmentation Methods in Medical Imaging

Abstract

Image semantic segmentation based on deep learning attracted much attention in the field of computer vision in recent years. In the past few decades, the tremendous developments of medical imaging technology have made it possess an increasingly important role in diagnosis. Therefore, the image semantic segmentation method based on the deep learning algorithm is widely considered by many researchers in the medical image processing field. In this paper, first of all, we introduced some semantic segmentation methods based on deep learning, including point cloud data-based methods, Recurrent Neural Network (RNN) based methods, attention mechanism-based methods, transformer-based methods, and Generative Adversarial Network (GAN) based methods. Then, the characteristics of each method and the problems improved by each method in the field of medical imaging are shown respectively, for example, point cloud data can better express the information of biological tissue structure to improve the capacity of processing global contextual dependencies, attention mechanism assists the model quickly discover key information of input images, transformer improve the effect of segmentation tasks, and GAN has superior data generation capacity that they have excellent performance in generating realistic-looking images. And we summarize the performance of each method in the medical dataset by distinct metrics. Finally, some problems of theory, architecture and application in future research were discussed.