CN-121999490-A - Glioblastoma patient survival prediction method and system

CN121999490ACN 121999490 ACN121999490 ACN 121999490ACN-121999490-A

Abstract

The invention provides a glioblastoma patient lifetime prediction method and a glioblastoma patient lifetime prediction system, and relates to the technical field of intelligent medical treatment; the method comprises the steps of screening a high-information-content image block based on information entropy, executing a dendritic calculation rule and a structural plasticity rule through a dendritic nerve calculation model by using a dendritic nerve capable of being learned in the image block, extracting global and local features from the image block, fusing the features through a feature fusion nerve to obtain depth features, and finally predicting the total lifetime through an output nerve. According to the invention, through simulating layering calculation and plasticity of the neuron dendrites, a double-layer-level feature learning model capable of integrating global context and local detail of a full-slice image is constructed, the technical problems of insufficient feature integration and limited model calculation structure of the existing prediction method are effectively solved, and more comprehensive and accurate prediction of the survival time of a glioblastoma patient is realized.

Inventors

YIN XIU
XUE JIE
LIU XIYU

Assignees

巢湖学院
山东师范大学

Dates

Publication Date: 20260508
Application Date: 20260128

Claims (10)

1. A method for predicting survival of a glioblastoma patient, comprising: acquiring a digital full-slice image of a glioblastoma patient, and encoding the digital full-slice image into a numerical matrix; Screening a plurality of image blocks with highest information content from the numerical matrix based on information entropy; Inputting the image block into a dendritic morphology nerve calculation model for processing, wherein the model comprises at least one learnable dendritic neuron, each neuron is internally provided with a dendritic tree structure with multiple layers of branches, each dendritic branch is used as an independent calculation unit, and global features and local features are extracted from the image block by executing preset communication rules; fusing the global features and the local features through feature fusion neurons in the model to generate fused depth features; And inputting the fused depth characteristics into an output neuron, and calculating to obtain the predicted value of the total lifetime of the patient.
2. The method of claim 1, wherein the predetermined communication rules include a dendrite calculation rule and a structural plasticity rule: The dendrite calculation rule is , wherein, As a branch variable of the dendrite, Is a learnable linear or nonlinear function; the structural plasticity rule is that , wherein, The weight of the value in the interval of [0,1] is used for dynamically adjusting the dendritic connection.
3. The method of claim 2, wherein the screening the plurality of image blocks with the highest information content from the numerical matrix based on information entropy comprises: And calculating the information entropy values of the image areas according to the dendrite calculation rule, and selecting a preset number of areas with the highest information entropy values as the image blocks.
4. The method of claim 1, wherein the feature fusion neuron generates a fused depth feature by a rule that Firstly according to Generating a attention profile According to And performing feature weighting and splicing, wherein, And The global features and the local features are respectively, Representing a matrix multiplication of the number of bits, Representing an element-by-element multiplication, Representing a stitching operation.
5. The method of claim 1, wherein the dendritic morphology nerve computation model further comprises a self-supervised learning architecture including an encoder for performing multi-level downsampling, feature extraction, and attention enhancement on the image block, and a decoder for reconstructing an image from depth features output by the encoder and by minimizing a mean square error loss between the reconstructed image and an original image To optimize model parameters According to An update is performed, wherein, Is the learning rate.
6. The method of claim 1, wherein the computational rules of the output neurons are Obtaining a preliminary predicted value , wherein, For the depth features after the fusion, And Is a learnable parameter; based on the preliminary predicted value And true lifetime According to the loss function Calculating a predicted loss, the total lifetime predicted value being , wherein, Is the number of patients.
7. A glioblastoma patient survival prediction system, comprising: the image acquisition and encoding unit is used for acquiring a digital full-slice image of a glioblastoma patient and encoding the digital full-slice image into a numerical matrix; the image block screening unit is used for screening a plurality of image blocks with highest information content from the numerical matrix based on information entropy; The feature extraction unit is used for inputting the image block into a dendritic morphology nerve calculation model for processing, the model comprises at least one learnable dendritic neuron, a dendritic tree structure with multiple layers of branches is arranged in each neuron, each dendritic branch is used as an independent calculation unit, and global features and local features are extracted from the image block by executing preset communication rules; the feature fusion unit is used for fusing the global features and the local features through the feature fusion neurons in the model to generate fused depth features; and the lifetime prediction unit is used for inputting the fused depth characteristics into an output neuron, and calculating to obtain the total lifetime predicted value of the patient.
8. An electronic device comprising a memory, a processor and a computer program stored on the memory and running on the processor, wherein the processor implements the steps of the glioblastoma patient lifetime prediction method according to any one of claims 1 to 6 when the program is executed by the processor.
9. A computer readable storage medium having stored thereon a computer program, wherein the program when executed by a processor implements the steps of the glioblastoma patient survival prediction method according to any one of claims 1 to 6.
10. A computer program product comprising software code, wherein a program in the software code performs the steps of the glioblastoma patient survival prediction method according to any one of claims 1 to 6.

Description

Glioblastoma patient survival prediction method and system Technical Field The invention relates to the technical field of intelligent medical treatment, in particular to a method and a system for predicting the survival time of glioblastoma patients. Background Glioblastoma (GBM) is the most invasive primary brain tumor, has the characteristics of high morbidity, high mortality and diversified clinical manifestations, has a five-year survival rate of only about 5% after diagnosis of patients, and has extremely poor prognosis, and its severity and complexity make it the focus of the field of neurooncology. The accurate total survival time prediction model is constructed, and is important for making an individual treatment scheme and improving disease prognosis. Currently, computational pathology based on deep learning is working on predicting cancer patient survival from digital whole-slice images (WSI). Typical automated survival prediction methods generally involve three key steps, image block sampling, block-level feature extraction, and slice-level characterization learning. However, the existing model architecture has the following problems that on one hand, a mainstream feature extraction model, such as a multi-instance learning framework, a multi-scale fusion network and the like, is limited by local visual field, global context information and local cell nucleus details of a full-slice image are difficult to integrate effectively, and on the other hand, a novel computational model which is inspired by biology, such as a pulse neural (SN) membrane system, has parallel computational advantages, but models neurons as simple computational units, and ignores strong and layered computational potential of a biological neuron dendrite structure. This results in the existing model not being sufficiently comprehensive and accurate for capturing complex pathological morphological features for accurate, individualized life-time predictions. Disclosure of Invention The invention provides a glioblastoma patient lifetime prediction method and a glioblastoma patient lifetime prediction system, which aim to solve the technical problems that the existing prediction model in the background technology is difficult to integrate global and local image information and can not fully utilize the internal calculation structure of neurons. The method is characterized in that a dendritic form nerve calculation model is introduced, complex feature processing is realized in a single neuron by simulating layering calculation and plasticity of biological neuron dendrites, and a double-layer learning framework integrating global and local features is constructed, so that more comprehensive and accurate prediction of the survival time of GBM patients is realized. To achieve the above object, a first aspect of the present invention provides a method for predicting survival of a glioblastoma patient, comprising: acquiring a digital full-slice image of a glioblastoma patient, and encoding the digital full-slice image into a numerical matrix; Screening a plurality of image blocks with highest information content from the numerical matrix based on information entropy; Inputting the image block into a dendritic morphology nerve calculation model for processing, wherein the model comprises at least one learnable dendritic neuron, each neuron is internally provided with a dendritic tree structure with multiple layers of branches, each dendritic branch is used as an independent calculation unit, and global features and local features are extracted from the image block by executing preset communication rules; fusing the global features and the local features through feature fusion neurons in the model to generate fused depth features; And inputting the fused depth characteristics into an output neuron, and calculating to obtain the predicted value of the total lifetime of the patient. Further, the preset communication rules include a dendrite calculation rule and a structural plasticity rule: The dendrite calculation rule is , wherein,As a branch variable of the dendrite,Is a learnable linear or nonlinear function; the structural plasticity rule is that , wherein,The weight of the value in the interval of [0,1] is used for dynamically adjusting the dendritic connection. Further, the screening the plurality of image blocks with the highest information content from the numerical matrix based on the information entropy comprises the following steps: And calculating the information entropy values of the image areas according to the dendrite calculation rule, and selecting a preset number of areas with the highest information entropy values as the image blocks. Further, the feature fusion neuron generates a fused depth feature by the following ruleFirstly according toGenerating a attention profileAccording toAnd performing feature weighting and splicing, wherein,AndThe global features and the local features are respectively,Representing a matrix m