深度学习脑肿瘤MRI图像分类研究进展

Research progress of deep learning brain tumor MRI image classification

大数据环境下肿瘤病例为肿瘤的临床诊断提供了庞大的数据资源,同时人工智能技术的发展促进深度学习应用水平不断提升,推动肿瘤MRI图像的快速、精准分类进入深度学习时代。本文主要分为以下四个部分,第一部分针对目前主流的深度学习MRI图像分类模型:卷积神经网络、深度信念网络、深度残差网络、Vision Transformer展开综述。首先,阐述了各模型的历史沿袭、最初针对的问题及主要思想;其次,概括了模型的网络架构并探讨其在MRI图像分类上的最新应用;然后,分析了模型的特点、目前存在的局限及各自发展趋势。第二部分论述了一些影响分类性能的关键因素;第三部分提出了一些广泛使用的性能增强技术;文章最后讨论了深度学习分类MRI图像在临床实践中面临的主要限制,并对未来研究方向进行展望。本文的结果可为研究人员提供一个全面的比较,以及各种深度学习模型的有效性,有望促进脑肿瘤研究的进展。

In the big data environment,tumor cases provide huge data resources for the clinical diagnosis of tumors.Meanwhile,the development of artificial intelligence technology promotes the continuous improvement of the application level of deep learning and promotes the rapid and accurate classification of tumor images in the era of deep learning.This paper is mainly divided into the following four parts.The first part reviews the current mainstream deep learning MRI image classification models:convolutional neural network,deep belief network,deep residual network,and Vision Transformer.Firstly,the historical lineage,the initial problems,and the main ideas of each model are described.Secondly,the network architecture of the model is summarized and its latest application in MRI image classification is discussed.Then,the characteristics,limitations,and development trends of the models are analyzed.The second part discusses some key factors that affect classification performance.In the third part,some widely used performance enhancement techniques are proposed.Finally,the main limitations of deep learning classification of MRI images in clinical practice are discussed,and future research directions have been prospected.The results presented here can provide researchers with a comprehensive comparison as well as the effectiveness of various deep learning models,which is expected to promote the progress of brain tumor research.