Semantic Segmentation of Cerebral Cortex Based On Pointnet

Yanting Sang

Abstract: The cerebral cortex is the highest level center for regulating and controlling body movement. The semantic segmentation of the cerebral cortex is of great significance for the study of the neural structure of the brain and the early diagnosis and treatment of brain diseases. At present, most of the existing cerebral cortex segmentation methods are based on the brain structure segmentation based on MRI images, and then extract the cortical information as the final segmentation result. This method has problems such as large amount of data and low segmentation efficiency, and the accuracy is easily affected by the brain. Influence of internal structure segmentation results. In response to this problem, this paper directly takes the cerebral cortex data as the research object, and on the basis of fully considering the characteristics of the cerebral cortex point cloud collection data, the PointNet segmentation network is used to perform semantic segmentation of the cerebral cortex. At the same time, in view of the imbalance of data categories in the dataset, a sample balance mechanism is introduced to improve the loss function of the segmentation model, thereby improving the contribution of smaller categories in the dataset to the segmentation results. The experimental results show that the PointNet network can well solve the problem of semantic segmentation of the cerebral cortex, and the use of the sample equalization mechanism can further improve the accuracy of small category partitioning and improve the segmentation accuracy.

Keywords: Cerebral cortex, semantic segmentation, PointNet, sample equalization