Artificial intelligence-based cerebral vascular disease dynamic prediction model construction method and system

By constructing a sparse topological graph network model and employing pruning and masking mechanisms to dynamically update the cerebrovascular disease prediction model, the problems of insufficient clinical interpretability and path drift in existing models are solved, achieving efficient and accurate dynamic prediction of cerebrovascular diseases.

CN122245703APending Publication Date: 2026-06-19TONGJI HOSPITAL ATTACHED TO TONGJI MEDICAL COLLEGE HUAZHONG SCI TECH
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
TONGJI HOSPITAL ATTACHED TO TONGJI MEDICAL COLLEGE HUAZHONG SCI TECH
Filing Date
2026-03-06
Publication Date
2026-06-19

AI Technical Summary

Technical Problem

Existing cerebrovascular disease prediction models suffer from insufficient clinical interpretability and path drift issues in long-term disease progression. They lack clear medical logical constraints and dynamic correction mechanisms, resulting in prediction results that lack intuitive basis and accuracy.

Method used

A graph network model with multiple consecutive time slices is constructed. Based on a sparse topology, the weights of the connecting edges are dynamically updated through pruning and masking mechanisms. Only the unique evolution path of the current cycle is retained, and other evolution branches that are not predetermined facts are hidden. A sparse penalty term and a Top-K filtering mechanism are introduced to eliminate redundant paths.

Benefits of technology

This approach enhances the clinical interpretability of the model and improves the accuracy of long-term disease course prediction. By eliminating noise interference through a sparse topology model, it ensures that the prediction process conforms to medical logic and improves the accuracy of personalized disease course prediction.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122245703A_ABST
    Figure CN122245703A_ABST
Patent Text Reader

Abstract

This invention relates to the field of neural network construction technology, and more particularly to a method and system for constructing a dynamic prediction model for cerebrovascular diseases based on artificial intelligence. The invention first constructs a graph network model with continuous time slices and defines multimodal feature nodes. Based on the evolutionary correlation of historical data, pruning and sparsity constraints are applied to the fully connected matrix between layers to construct a sparse topology model. Subsequently, fact evolution nodes are activated based on patient clinical data, and reverse path backtracking is performed with these nodes as endpoints. A masking mechanism is used to block non-predetermined fact branches, locking the unique historical path. Finally, this path is transformed into a context vector and injected into subsequent layers to achieve model state reconstruction and dynamic weight updates. This invention, through a global pruning-path locking-dynamic reconstruction mechanism, effectively filters clinical noise in disease progression, solves the path drift problem of traditional time-series prediction, and significantly improves the accuracy of individualized predictions in the mid-to-late stages while greatly reducing the number of parameters.
Need to check novelty before this filing date? Find Prior Art