融合多模态多尺度磁共振成像的脑胶质瘤分割

Brain Glioma Segmentation with Multi-modal and Multi-scale Magnetic Resonance Imaging

为了实现脑胶质瘤小目标区域的精准分割,提出了融合多模态多尺度磁共振成像脑胶质瘤分割模型。通过多模态特征提取模块获取各模态的图像特征,增强网络对特征信息的复用性;利用多尺度特征融合模块学习不同尺度的关键特征,提升网络对小目标脑胶质瘤区域的特征识别能力;使用一种加权混合损失函数解决类不平衡问题。在BraTS(brain tumor segmentation)2019数据集上测试该模型,其中整个肿瘤、核心肿瘤和增强肿瘤的Dice系数分别为0.857、0.869和0.878,Hausdorff距离分别为2.543、1.583和1.526。实验结果表明,该模型可以有效提高脑胶质瘤小目标区域的分割精度。

To achieve precise segmentation of small target regions of glioma,a multi-modal and multi-scale MRI glioma seg-mentation model is proposed.The image features of each modality are obtained through the multi-modal feature extraction module,which enhances the reusability of the feature information by the network.The multi-scale feature fusion module is used to learn key features at different scales,and improve the feature recognition ability of the network for small target glioma regions.A weighted hy-brid loss function is used to address the class imbalance problem.The proposed model is tested on the BraTS(brain tumor segmenta-tion)2019 dataset,where the Dice scores of the whole tumor,tumor core,and enhancing tumor are 0.857,0.869 and 0.878,and Hausdorff distances are 2.543,1.583 and 1.526,respectively.The experimental results show that the model can effectively improve the segmentation accuracy of small target regions of glioma.