基于多模态MRI中深层语义和边缘信息融合的脑肿瘤分割 | 文献速递-深度学习肿瘤自动分割

Title

题目

Brain tumor segmentation based on the fusion of deep semantics and edge information in multimodal MRI

基于多模态MRI中深层语义和边缘信息融合的脑肿瘤分割

01

文献速递介绍

医学图像分割是医学图像处理领域的重要课题。其中，脑肿瘤分割旨在从图像中定位多种类型的肿瘤区域，对临床实践具有重要意义。由于磁共振成像（MRI）在提供软组织高分辨率解剖结构方面的良好能力，它在诊断和治疗脑肿瘤疾病中被广泛应用。为了获得准确分割所需的全面信息，通常需要在脑肿瘤分割中使用具有不同成像参数的多模态MRI扫描。常用的模态包括液体衰减反转恢复（FLAIR）、T1加权（T1）、对比增强T1加权（T1ce）和T2加权（T2）。不同模态的图像捕获了不同的病理信息，并且它们可以有效地相互补充，这在分割多种类型的脑肿瘤区域（如水肿（ED）、坏死和非增强肿瘤（NCR/NET）以及增强肿瘤（ET））中起着至关重要的作用。图1展示了用于脑肿瘤分割的多模态MRI的示例。为简单起见，只选择了整个扫描的一个切片。图1(a)显示了领域专家提供的地面真相（GT）分割标签。绿色、黄色和红色分别表示ED、ET和NCR/NET区域。从图1(b)到(e)可以看出，不同模态的特征差异显著。例如，FLAIR模态可以很好地捕获具有明显边缘或边界的ED区域，而T1ce模态在检测具有高对比度的肿瘤核心（即ET和NRC/NET的联合）方面更为有效。

Abstract

摘要

Brain tumor segmentation in multimodal MRI has great significance in clinical diagnosis and treatment. Theutilization of multimodal information plays a crucial role in brain tumor segmentation. However, most existingmethods focus on the extraction and selection of deep semantic features, while ignoring some features withspecific meaning and importance to the segmentation problem. In this paper, we propose a brain tumorsegmentation method based on the fusion of deep semantics and edge information in multimodal MRI, aimingto achieve a more sufficient utilization of multimodal information for accurate segmentation. The proposedmethod mainly consists of a semantic segmentation module, an edge detection module and a feature fusionmodule. In the semantic segmentation module, the Swin Transformer is adopted to extract semantic featuresand a shifted patch tokenization strategy is introduced for better training. The edge detection module isdesigned based on convolutional neural networks (CNNs) and an edge spatial attention block (ESAB) ispresented for feature enhancement. The feature fusion module aims to fuse the extracted semantic and edgefeatures, and we design a multi-feature inference block (MFIB) based on graph convolution to perform featurereasoning and information dissemination for effective feature fusion. The proposed method is validated on thepopular BraTS benchmarks. The experimental results verify that the proposed method outperforms a numberof state-of-the-art brain tumor segmentation methods.

多模态M