ParkMultiNet combines MRI and speech data for multimodal Parkinson’s disease classification

Early Parkinson’s disease (PD) classification requires jointly reasoning over structural imaging and behavioral signals, yet existing methods remain limited in modeling long-range spatial dependencies in 3D medical images and in effectively integrating heterogeneous modalities, making it difficult to fully exploit subtle structural abnormalities and cross-modal complementary information. To address this issue, we propose ParkMultiNet, a multimodal framework for PD classification that integrates MRI and speech information. ParkMultiNet adopts a four-stage hierarchical architecture and contains three key modules: a multi-view spatial attention (MVSA) module for modeling long-range 3D spatial dependencies through decoupled convolutional paths, a group-enhanced channel recalibration (GECR) module for suppressing feature redundancy and improving modality-aware feature refinement, and a cascaded atrous spatial pyramid module (CASPM) for extracting multi-scale contextual features while maintaining parameter efficiency. Experiments were conducted on PubPDdataset, a pseudo-paired multimodal dataset constructed from multiple public data sources, and on PativePDdataset, an independent clinical dataset evaluated in a label-consistent pseudo-paired multimodal setting. Results show that ParkMultiNet achieves the highest observed mean accuracy among the compared methods in multimodal classification. In particular, it attains an accuracy of 0.9385 and an F1-score of 0.9457 on PubPDdataset. On PativePDdataset, its observed mean accuracy was 0.8417. Further unimodal comparisons and ablation studies verify the effectiveness of multimodal integration and the complementary contributions of MVSA, GECR, and CASPM. These results indicate that ParkMultiNet provides an effective multimodal solution for PD classification.

Authors

Institutions

Publication Details

Journal
Scientific Reports
Published
2026-11-14
DOI
https://doi.org/10.1038/s41598-026-77542-1
Primary Topic
Parkinson's Disease Mechanisms and Treatments
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

ParkMultiNet combines MRI and speech data for multimodal Parkinson’s disease classification

Xinling Yang, Lu Hao, Guoyong Jia, PeiShan Li et al.
Scientific Reports
Parkinson's Disease Mechanisms and Treatments
article

ParkMultiNet combines MRI and speech data for multimodal Parkinson’s disease classification

Xinling Yang, Lu Hao, Guoyong Jia, PeiShan Li, Zihao Wang
article en

Abstract

Early Parkinson’s disease (PD) classification requires jointly reasoning over structural imaging and behavioral signals, yet existing methods remain limited in modeling long-range spatial dependencies in 3D medical images and in effectively integrating heterogeneous modalities, making it difficult to fully exploit subtle structural abnormalities and cross-modal complementary information. To address this issue, we propose ParkMultiNet, a multimodal framework for PD classification that integrates MRI and speech information. ParkMultiNet adopts a four-stage hierarchical architecture and contains three key modules: a multi-view spatial attention (MVSA) module for modeling long-range 3D spatial dependencies through decoupled convolutional paths, a group-enhanced channel recalibration (GECR) module for suppressing feature redundancy and improving modality-aware feature refinement, and a cascaded atrous spatial pyramid module (CASPM) for extracting multi-scale contextual features while maintaining parameter efficiency. Experiments were conducted on PubPDdataset, a pseudo-paired multimodal dataset constructed from multiple public data sources, and on PativePDdataset, an independent clinical dataset evaluated in a label-consistent pseudo-paired multimodal setting. Results show that ParkMultiNet achieves the highest observed mean accuracy among the compared methods in multimodal classification. In particular, it attains an accuracy of 0.9385 and an F1-score of 0.9457 on PubPDdataset. On PativePDdataset, its observed mean accuracy was 0.8417. Further unimodal comparisons and ablation studies verify the effectiveness of multimodal integration and the complementary contributions of MVSA, GECR, and CASPM. These results indicate that ParkMultiNet provides an effective multimodal solution for PD classification.

Scientific Reports
Xinjiang Medical University (CN), Second Affiliated Hospital of Xinjiang Medical University (CN)
Parkinson's Disease Mechanisms and Treatments
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon