A method and system for Parkinson's disease diagnosis based on missing perception hybrid expert network modality reconstruction

By employing a missing data-aware hybrid expert network modality reconstruction method, dynamically selecting reconstruction strategies and introducing modality availability tokens, the accuracy problem of Parkinson's disease diagnosis under multimodal data missing conditions is solved, achieving efficient diagnosis in modal incomplete environments.

CN123120263APending Publication Date: 2026-11-17ZHEJIANG UNIV +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611287265.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-24
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing Parkinson's disease diagnostic methods suffer from decreased diagnostic accuracy and generalization ability when multimodal data is missing, and simple imputation strategies are prone to introducing noise, making it impossible to effectively utilize existing modal information for accurate diagnosis.

Method used

A missing modality reconstruction method using a hybrid expert network is adopted. The missing modality features are reconstructed and cross-modality features are fused by dynamically selecting the most suitable reconstruction strategy through a missing-aware gating network and an expert network, combined with semantic contrast loss and modality availability token.

Benefits of technology

Under incomplete modal conditions, accurate diagnosis of Parkinson's disease was achieved, improving diagnostic accuracy and stability while maintaining the model's excellent performance in real clinical settings.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123120263A_ABST
    Figure CN123120263A_ABST
Patent Text Reader

Abstract

This invention discloses a Parkinson's disease diagnostic method and system based on missing-aware hybrid expert network modality reconstruction, relating to the field of medical disease diagnosis. The method includes: preprocessing multimodal data and extracting the true features of each modality; if some modality data is missing, creating virtual features for the missing modality, and constructing low-dimensional shared features using these virtual features and the true features of existing modalities; reconstructing features of the missing modality based on the low-dimensional shared features and using a missing-aware hybrid expert network and semantic contrast loss to obtain reconstructed features of the missing modality; fusing the reconstructed features of the missing modality with the true features of existing modalities to obtain full-modality features; introducing a modality availability token to explicitly represent currently available modality information; and fusing the modality availability token with the full-modality features for Parkinson's disease diagnosis prediction. This method requires only a small number of parameters to maintain high diagnostic accuracy in modality-missing scenarios and is suitable for auxiliary diagnosis of clinical Parkinson's disease.
Need to check novelty before this filing date? Find Prior Art