Gradient-assisted deep model for brain tumor segmentation by multi-modality MRI volumes

Abstract

In clinical diagnosis, doctors make treatment strategies through key factors such as the location, shape and size of tumors. With the development of computer technology, the advantages of deep learning in medical image segmentation are increasingly prominent. Gliomas have the characteristics of diffuse infiltration, blurred boundary and swelling of brain tissue in affected areas, which makes it challenging to accurately segment brain tumors near the intersection area. The existing algorithms have achieved great results in segmenting tumors in terms of gray information. However they ignored the gradient of tumor boundary areas. The complexity of multi-modality MRI and the vast differences between brain tumor areas make it difficult to segment brain tumors efficiently and accurately. To solve the above problems, we propose a gradient-assisted multi-category brain tumor segmentation method(GAM-Net). GAM-Net includes three branches: (1) double convolutional encoder, which could capture rich features from multi-modality MRI; (2) gradient extraction branch, which could generate gradient features to assist area segmentation; and (3) gradient-driven decoder which could provide fusion contour information and encoding features effectively. We evaluated the effectiveness of the proposed algorithm on BraTS2020 dataset, of which 295 cases are used as training sets and 74 cases as test sets. Finally, the Dice Similarity Coefficients (DSC) of the proposed algorithm in whole tumor (WT), tumor core (TC), and enhanced tumor (ET) are 0.8991, 0.8402, and 0.7580 respectively. Average DSC reaches 0.8324. Experimental results show that GAM-Net can be successfully applied to segment brain tumors and thus helpful in diagnosis and treatment.