Prior boundary-fused brain tumor segmentation method

Abstract

The invention discloses a prior boundary-fused brain tumor segmentation method, and aims to overcome the defects that the generated brain tumor segmentation boundary is rough and the tumor is easy to reconstruct due to the fact that an existing convolutional network cannot fully utilize global image information. An optimal boundary of a tumor truth value is obtained from tumor prior knowledge, an optimal boundary generation network is constructed, an optimal boundary is added to a 3D U-net network of multiple down-sampling channels to carry out weight distribution and boundary enhancement on each layer of the network, and the similarity between a generated tumor edge and a tumor truth value edge is used as a loss item to be added to an original loss function to improve the accuracy of edge segmentation. According to the method, the prior knowledge is fused into the network and the loss function by using the tumor information of different modal nuclear magnetic images through multiple down-sampling channels, so that the comprehensiveness of tumor information utilization and the accuracy of tumor edge segmentation are improved.

Description

technical field [0001] The invention relates to medical image processing, in particular to a brain tumor MRI image segmentation method. Background technique [0002] Brain tumor image segmentation based on medical imaging is an application of computer vision in medical image processing, which mainly uses image features to separate tumor regions from other tissue regions. Reliable brain tumor segmentation is critical for accurate medical diagnosis and subsequent treatment. Since manual segmentation of brain tumors requires expert manual labeling and screening of tissues, which is a time-consuming, expensive, and subjective task, practical automated methods are highly desired. But because brain tumors are highly heterogeneous in location, shape, and size, developing automatic segmentation methods has been a daunting task for decades. [0003] Scholars have proposed a variety of segmentation methods based on brain tumor MRI images. The most typical traditional segmentation m...

AI Technical Summary

Problems solved by technology

[0004] In order to overcome the problem that the existing convolutional network cannot make full use of the global image information, resulting in rough brain tumor segmentation boundaries and common spurious problems in reconstructed tumors, the present invention proposes a brain tumor segmentation method that fuses prior boundaries, and utilizes multiple downsampling channels The tumor information of different modal MRI images, the prior knowledge is integrated into the network and loss function, which improves the comprehensiveness of tumor information utilization and the accuracy of tumor edge segmentation

Images