A training method of a venous thromboembolism risk prediction model

By combining neural differential equations and a bi-branch competitive risk prediction network, the problem of information loss in venous thrombosis risk prediction models under inconsistent sampling frequencies and data sparsity was solved, enabling early and accurate risk warning and dynamic intervention, thus improving the accuracy of risk assessment and the efficiency of clinical decision-making.

CN122136006BActive Publication Date: 2026-07-14THE FIRST AFFILIATED HOSPITAL ZHEJIANG UNIV COLLEGE OF MEDICINE
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
THE FIRST AFFILIATED HOSPITAL ZHEJIANG UNIV COLLEGE OF MEDICINE
Filing Date
2026-05-07
Publication Date
2026-07-14

AI Technical Summary

Technical Problem

Existing venous thrombosis risk prediction models are prone to losing nonlinear change information when dealing with inconsistent sampling frequencies or sparse data of physiological data, resulting in low accuracy of risk warnings and difficulty in providing accurate risk warnings in the early stages of thrombosis formation.

Method used

We employ neural constant differential equations for continuous state reconstruction, combined with a bi-branch competitive risk prediction network, to construct a time-sensitive and cost-sensitive joint loss function, and output a dynamic intervention window matrix to assist clinical decision-making.

Benefits of technology

It significantly improves the accuracy and continuity of risk warnings, providing precise risk warnings in the early stages of thrombosis, reducing the medical costs of false alarms and the health risks of underreporting, and enabling on-demand allocation of medical resources and monitoring accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122136006B_ABST
    Figure CN122136006B_ABST
Patent Text Reader

Abstract

The present application relates to the field of artificial intelligence and medical data processing, in particular to a training method of a venous thrombosis risk prediction model, comprising a multi-modal feature extraction step: acquiring multi-modal non-aligned time series data, and extracting discrete and continuous sampling features; a continuous state reconstruction step: inputting the features into a neural differential equation, filling in observation blanks, and dynamically generating a continuous hidden state and an evolution rate vector; a competitive risk prediction step: combining the state vector, using a double-branch network to simulate the probability evolution of thrombosis events and competitive events respectively, and generating a cumulative baseline hazard rate and a competitive event hazard rate; a cost-sensitive optimization step: introducing the idea of game theory to construct a time-sensitive and cost-sensitive joint loss function, balancing timeliness and false alarm cost, and completing parameter updating; and a dynamic intervention output step: using the model to infer and output a dynamic intervention window matrix; the present application effectively overcomes the problem of providing accurate early warning in the early stage of thrombosis.
Need to check novelty before this filing date? Find Prior Art