Risk prediction method based on multi-source medical data

By combining a multi-stage feature selection strategy with multiple evaluation methods and machine learning models, the problem of inefficient feature selection in multi-source heterogeneous medical data is solved, the screening of key features and the close integration of the model are achieved, and the prediction performance and stability are improved.

CN120674076APending Publication Date: 2025-09-19DALIAN UNIV OF TECH
View PDF 4 Cites 3 Cited by

Patent Information

Application Number
CN202510806430.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-06-17
Publication Date
2025-09-19

AI Technical Summary

Technical Problem

When processing multi-source heterogeneous medical data, existing technologies have low feature selection efficiency, fail to fully capture complex associations, lack close correlation with subsequent model performance, lack robustness, and have difficulty screening out key features, especially in clinical, laboratory, and cardiac MRI data.

Method used

A multi-stage feature selection strategy is adopted, combining filtering, wrapping and embedding methods for preliminary feature voting. Through feature ranking integration and stepwise forward selection, a feature subset for predicting specific medical events or disease risks is screened out, and combined with machine learning models for training and interpretation.

Benefits of technology

It improves the effectiveness and robustness of feature screening, can screen out highly relevant feature combinations from multi-source heterogeneous data, enhances the interpretability and predictive performance of the model, is suitable for multi-source heterogeneous data, reduces model complexity and improves stability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120674076A_ABST
    Figure CN120674076A_ABST
Patent Text Reader

Abstract

The invention belongs to the technical field of medical data processing and health assessment, and discloses a multi-source medical data-based risk prediction method, which comprises the steps of multi-source medical data acquisition, data preprocessing, feature selection, model training, risk prediction and model explanation. According to the method, by adopting a systematic multi-stage feature selection strategy, a specific key feature combination highly related to a specific medical event or disease can be screened out from multi-source heterogeneous medical data including clinical data, laboratory data, heart MRI (Magnetic Resonance Imaging) and the like; the screened feature subsets can be used for constructing an interpretable machine learning model, and through combination with SHAP and other model interpretation technologies, clinicians are helped to understand prediction logic, the credibility of results is enhanced, and more valuable reference information is provided for individualized clinical decision and intervention.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Ovarian cancer molecular typing prediction system

    CN110766678A

  • A Feature-Based Method for Selecting Main Symptoms in Traditional Chinese Medicine

    CN113707293B

  • A functional connectivity matrix processing system and apparatus based on feature selection using filtering method

    CN115147417B

  • Reinforcement learning intrusion detection method and system based on dynamic network feature screening

    CN119696934A

Cited By

  • Rare disease risk screening model training system and method, screening system and medium

    CN121075621A

  • Complication risk prediction system and device based on agent workflow series scheduling

    CN121460192A

  • Aircraft safety state assessment method based on multi-source data analysis

    CN121561752A