Cascaded cavity convolutional network brain tumor segmentation method with attention mechanism

Abstract

The invention relates to a cascaded cavity convolutional network brain tumor segmentation method with an attention mechanism. The method comprises the following steps: data preprocessing; a network structure is built, and the method comprises the following steps that: a cascaded cavity convolutional network with an attention mechanism is built, a three-level cascaded framework is adopted, multipletypes of segmentation tasks are simplified into three second-class segmentation tasks, and the three-level segmented networks are respectively W-Net, T-Net and E-Net and are respectively used for segmenting a whole brain tumor (WT) region, a tumor nucleus (TC) region and an enhanced tumor nucleus (ET) region; segmentation is carried out from the axial direction, the sagittal direction and the coronal direction in each stage, then averaging is carried out in segmentation results in the three directions, and an accurate segmentation result is obtained; a network structure of each level in a three-level cascaded framework of cascaded hole convolution with an attention mechanism is a full convolutional network structure for encoding and decoding, and is divided into four parts, namely an encoder, a decoder, a skip layer structure and multi-layer feature map fusion.

Description

technical field [0001] The invention relates to the field of image processing, especially for three-dimensional medical brain tumor image segmentation. Background technique [0002] Brain tumors are intracranial tumors with high lethality. According to histological heterogeneity and tumor aggressiveness, brain tumors can be divided into high glial tumors (HGG) and low glial tumors (LGG). Brain tumors It can be further divided into edema area, tumor nuclear area, enhancing tumor nuclear area, non-enhancing tumor nuclear area, and necrotic area. The four modality images of brain tumor magnetic resonance (MR): T1, T1ce, T2 and FLAIR focus on different tumor areas and can provide complementary information for each other. Brain tumor segmentation is to separate different tumor regions in brain images, which is very important for patient disease assessment, treatment plan formulation and follow-up observation research. However, it is time-consuming and labor-intensive to rely on...

AI Technical Summary

Problems solved by technology

However, manual segmentation by doctors is time-consuming and labor-intensive, and it is prone to errors after a long period of manual labeling. Doctors with different experience will have different segmentation results. Therefore, an automatic and accurate brain tumor segmentation method is needed.

Images