Brain Tumor Segmentation Using Multi-Cascaded Convolutional Neural Networks and Conditional Random Field

Abstract

Accurate segmentation of brain tumor is an indispensable component for cancer diagnosis and treatment. In this paper, we propose a novel brain tumor segmentation method based on multicascaded convolutional neural network (MCCNN) and fully connected conditional random fields (CRFs). The segmentation process mainly includes the following two steps. First, we design a multi-cascaded network architecture by combining the intermediate results of several connected components to take the local dependencies of labels into account and make use of multi-scale features for the coarse segmentation. Second, we apply CRFs to consider the spatial contextual information and eliminate some spurious outputs for the fine segmentation. In addition, we use image patches obtained from axial, coronal, and sagittal views to respectively train three segmentation models, and then combine them to obtain the final segmentation result. The validity of the proposed method is evaluated on three publicly available databases. The experimental results show that our method achieves competitive performance compared with the state-of-the-art approaches.INDEX TERMS Brain tumor segmentation, convolutional neural network, multi-cascaded convolutional neural network, conditional random field, multi-modality.

IOUC-3DSFCNN: Segmentation of Brain Tumors via IOU Constraint 3D Symmetric Full Convolution Network with Multimodal Auto-context

“…In recent years, deep learning models, with convolution neural networks (CNNs), have revolutionized computer vision through end-to-end learning, i.e., integrating automated features learning from large-scale raw image data with supervised classifier, attracting increasing interest in brain tumor segmentation, which can achieve the state-of-art segmentation result 6 . For instance, Pereira et al 5 proposed a deep CNN-based automatic glioma segmentation model with small convolutional kernels; Hu et al 16 combined multi-cascaded CNN with fully connected conditional random fields (CRFs) for the brain tumor segmentation.…”

“…Although U-Net-based segmentation models can achieve relatively good segmentation performance in many biological tissue segmentation tasks, it is still intractable in the accurate segmentation of complex brain tumors. As concluded in many reports 7 that brain tumor segmentation from MR images, especially the gliomas segmentation, is still a challenging task 6,16,18 .…”

“…Since these methods cannot make a full use of 3D context information. There is usually a lack of reliability and accuracy in object segmentation from volumetric data 16 . In addition, the direct use of 3D CNN (such as 3DU-Net) also shows limited adaptability when the network is not deep enough 20 , whereas the deep networks are computation intensive and prone to over-fitting.…”

“…To make a comprehensive comparison with the latest brain tumor segmentation methods, the experimental prediction results of some state-of-the-art models are listed in Table 6, where the experimental results are all the best reported values in theiroriginatedliterature. As listed in Table 6, Hu et al 16 , Pereira et al 5 , Razzak et al 45 , Yang et al 46 , Zhao et al 21 and Xue et al 47 all used 2D CNNs to perform brain tumor segmentation. Sun et al 48 used a kind of 3DU-Net-like segmentation model.…”

Image classification-based brain tumour tissue segmentation

“…A novel method was presented in [14], which designed multi-cascade CNNs to take care of several local pixel dependencies and multi-scale features of 3D MRI images. The output results of the CNNs were refined using conditional random fields.…”