目的 脑肿瘤分割是医学图像处理中的一项重要内容,其目的是辅助医生做出准确的诊断和治疗,在临床脑部医学领域具有重要的实用价值。核磁共振成像(MRI)是临床医生研究脑部组织结构的主要影像学工具,为了使更多研究者对MRI脑肿瘤图像分割理论及其发展进行探索,本文对该领域研究现状进行综述。方法 首先总结了用于MRI脑肿瘤图像分割的方法,并对现有方法进行了分类,即分为监督分割和非监督分割；然后重点综述了基于深度学习的脑肿瘤分割方法,在研究其关键技术基础上归纳了优化策略；最后介绍了脑肿瘤分割(BraTS)挑战,并结合挑战中所用方法展望了脑肿瘤分割领域未来的发展趋势。结果 MRI脑肿瘤图像分割方法各有优缺点和其适用范围,研究者们一直致力于如何提高分割结果的精确度、模型的鲁棒性及运行效率。为了实现此目的本文分析总结了各类方法的优点、局限性。结论 MRI脑肿瘤图像分割领域的研究已经取得了一些显著进展,尤其是深度学习的发展为该领域的研究提供了新的思路。但由于脑肿瘤在大小、形状和位置方面的高度变化,以及脑肿瘤图像数据有限且类别不平衡,使得脑肿瘤图像分割仍是一个极具挑战的课题。由于分割过程缺乏可解释性和透明性,如何将全自动分割方法应用于临床试验,还需要进行深入的研究。
Review of tumor segmentation methods in brain MRI images
Li Qiang,Bai Kexin,Zhao Liu,Guan Xin(School of Microelectronics,Tianjin University)
Objective Brain tumor segmentation is an important part of medical image processing. Its purpose is to assist doctors to make accurate diagnosis and treatment. It has important practical value in clinical brain medicine. With the development of medical imaging, imaging technology plays an important role in the evaluation of the treatment of brain tumor patients, and can provide doctors with a clear internal structure of the human body. Magnetic resonance imaging (MRI) is the main imaging tool for clinicians to study the structure of brain tissue. Brain tumor MRI modalities include T1-weighted, contrast-enhanced T1-weighted, T2-weighted and liquid attenuation inversion recovery pulses. Different imaging modalities can provide complementary information to analyze brain tumors. These four kinds of images are usually combined to diagnose the location and size of brain tumors in the clinic. At present, due to the extensive application of MRI equipment in brain examination, a large number of brain MRI images will be generated in the clinic. It is impossible for doctors to manually annotate and segment all images in time, and manual segmentation of brain tumor tissue depends on the doctor"s personal experience. Therefore, how to segment brain tumors efficiently, accurately and automatically has become the focus of research. In recent years, significant advancement has been made in the study of brain tumor segmentation methods. In order to enable more researchers to explore the theory and development of brain tumor image segmentation on MRI, this paper reviews the current research status in this field. Method In this paper, the current semi-automatic and full-automatic segmentation methods based on MRI for brain tumors are divided into two categories: unsupervised segmentation and supervised segmentation. The difference between the two methods is whether or not to use the hand-labeled image data. Unsupervised segmentation is a non-priori image segmentation method based on computer clustering statistical analysis. The unsupervised methods are divided into threshold-based, region-based, pixel-based and model-based segmentation technologies according to the different segmentation principles. This paper briefly describes the unsupervised methods according to the above classification and summarizes the advantages and disadvantages of these methods. The main feature of supervised segmentation is making use of labeled image data. The segmentation process involves the model training section of using labeled data to learn the mapping from image features to labels and the test section of assigning labels to unlabeled data by the model. The supervised method is mainly based on the segmentation technology of pixel classification. Generally speaking, it includes the traditional machine learning and the method based on neural network. This paper briefly describes the common traditional machine learning methods and the method based on neural network used in brain tumor segmentation and summarizes the advantages and disadvantages of these methods. Then, the methods of brain tumor segmentation based on deep learning are mainly described. With the advance of artificial intelligence, deep learning especially the new technology represented by the Convolutional Neural Network (CNN) has been welcomed by showing the superior brain tumor segmentation results. Compared with traditional segmentation methods, CNN can automatically learn representative complex features directly from the data itself. Due to this characteristic, the research of brain tumor segmentation based on CNN mainly focuses on network structure design rather than image processing before feature extraction. This paper focuses on the structure of neural networks used in the field of brain tumor image segmentation, and summarizes the optimization strategies of deep learning. Finally, the challenge of brain tumor segmentation (BraTS) is introduced, and the future development trend of brain tumor segmentation is prospected in combination with the methods used in the challenge. The BraTS Challenge is a competition to evaluate MRI brain tumor segmentation methods. The BraTS challenge uses pre-operative MRI image data from multiple institutional brains to focus on the segmentation of gliomas. In addition, the BraTS challenge also predicts overall patient survival by combining radiological features with machine learning algorithms to determine the clinical relevance of this segmentation task. Result The segmentation methods of brain tumors on MRI have their own advantages, disadvantages and the scope of application. Researchers have been working on how to improve the accuracy of segmentation results, robustness of models and operational efficiency. In order to achieve this goal, this paper analyzes the advantages and disadvantages of various methods, the optimization strategy of deep learning and the future development trend. The optimization strategy of deep learning is: In the aspect of image data, use data enhancement techniques such as flipping, scaling and cropping to increase the amount of training data and improve the generalization ability of the model; Introduce cascade framework to realize the segmentation of whole tumors, core tumors and enhanced tumors respectively by combining with the inclusion relationship of tumors in brain anatomical structure; Use an improved loss function to deal with image category imbalances. In terms of network structure, multi-scale and multi-channel strategies are adopted to make full use of image feature information; In the process of down-sampling, convolution operation is used instead of pooling, so image information can be further learned while reducing image information loss; Between the convolutional layers, the connection method of the jump connection is used to effectively solve the degradation problem of the deep network; In different cases, the appropriate standardization method, activation function and loss rate are selected to achieve better segmentation effect. This paper summarizes the development trend of brain tumor segmentation method by learning and arranging the method used in the BraTS challenge. Due to the diversification of MRI imaging modality, making full use of each modal image information can effectively improve the segmentation accuracy of brain tumors. So how to use multi-modality images reasonably will become a research hotspot; The method based on deep learning is outstanding in the field of brain tumor segmentation and has become a hot research direction; The defect of machine learning algorithm will lead to inaccurate segmentation results of brain tumors. It is a popular trend to improve the original method or combine various methods effectively. Conclusion Some remarkable progress has been made in the field of brain tumor image segmentation on MRI. Especially the development of deep learning provides new ideas for the research in this field. However, brain tumor image segmentation is still a challenging subject because brain tumors vary obviously in size, shape and position. And brain tumor image data is limited and the categories are not balanced. Due to the lack of interpretability and transparency in the segmentation process, how to apply the fully automated segmentation method to clinical trials still needs further research.