Segmentation method of PET-CT multimodal nasopharyngeal carcinoma image based on hypergraph model

Abstract

The present invention relates to a PET-CT multimodal nasopharyngeal carcinoma image segmentation method based on a hypergraph model, comprising: extracting grayscale information and position information of pixels in a nasopharyngeal tumor image to construct a data set, and constructing a data set according to the data set Sparsely represent the model and solve it to obtain the reconstruction coefficient matrix and construct the hyperedge, use the Gaussian kernel model to calculate the similarity of data samples as the hyperedge weight value, solve the hyperedge order and vertex order to construct the hypergraph Laplacian matrix; then the nasopharynx The tumor image is marked and the label vector is obtained at the same time. A semi-supervised learning model is constructed according to the label vector, and then the optimal cutting vector is obtained by solving the least squares problem. Finally, the classification result is returned to the pixel level, that is, the segmentation of the tumor image is completed. The segmentation method of the present invention has higher segmentation accuracy than single modality, and at the same time, the hypergraph model based on the combination of sparse representation and Gaussian kernel has higher segmentation accuracy for nasopharyngeal tumor image data than other simple graph models or hypergraph models precision.

Description

technical field [0001] The invention belongs to the field of tumor image segmentation, and in particular relates to a PET-CT multimodal nasopharyngeal carcinoma image segmentation method based on a hypergraph model. Background technique [0002] With the development of medical imaging technology, medical image segmentation has gradually become an important research topic and has great significance in clinical applications. Especially in the processing of tumor images, accurate segmentation of tumor images can help doctors more accurately outline tumor contours to determine radiotherapy targets. However, most of the traditional tumor image segmentation methods are aimed at single-modal image data, which contains less information, so it is difficult to achieve high segmentation accuracy. [0003] Tumor is a malignant disease that endangers human health. Head and neck cancer is the fifth most common tumor in the world. Nasopharyngeal carcinoma, as the most common tumor in the ...

AI Technical Summary

Problems solved by technology

For example: use the K nearest neighbor algorithm to determine the connection of sample points to construct a hyperedge, and then obtain the weight value of the hyperedge based on the hyperedge. However, due to the fixed K value of this method, it is difficult to obtain ideal classification results when the sample distribution is uneven.

[0007] Nasopharyngeal carcinoma is one of the high-incidence tumors in my country, but currently there are few studies on image segmentation of nasopharyngeal tumors, and the results of related research results are not ideal. Research methods

Images