基于多尺度特征提取的MRI脑肿瘤图像分割研究

Research on MRI brain tumor image segmentation based on multi-scale feature extraction

针对磁共振成像(magnetic resonance imaging,MRI)脑部肿瘤区域误识别及肿瘤形状差异较大的问题,提出一种基于多尺度特征提取的MRI脑肿瘤图像分割方法。分割模型以U-Net为骨干网络,使用密集金字塔卷积(dense pyramidal convolution,DPC)提取多尺度特征,以适应不同尺寸肿瘤的分割,同时引入条状池化(strip pooling,SP),凭借其能捕获肿瘤中远距离区域的依赖关系,进一步加强对肿瘤图像的分割能力。提出的方法在Kaggle_3m数据集上进行了实验验证,实验结果表明该方法具有良好的脑部肿瘤分割性能,其中Dice相似系数、杰卡德系数分别达到了91.66%,84.38%。

Aiming at the problem of misrecognition of magnetic resonance imaging(MRI)brain tumor regions and large differences in tumor shape,a MRI brain tumor image segmentation method based on multi-scale feature extraction is proposed.The segmentation model uses U-Net as the backbone network and uses dense pyramid convolution(DPC)to extract multi-scale features to adapt to the segmentation of tumors of different sizes.At the same time,it introduces strip pooling(SP),which can capture the long-distance area of the tumor.Dependency,further strengthen the segmentation ability of tumor images.The proposed method has been experimentally verified on the Kaggle_3m data set.The experimental results show that the method has good brain tumor segmentation performance.The Dice similarity coefficient and the Jaccard coefficient have reached 91.66% and 84.38%,respectively.