A novel fusion architecture for detecting Parkinson’s Disease using semi-supervised speech embeddings

We introduce a framework for screening Parkinson’s disease (PD) using English pangram utterances. Our dataset includes 1306 participants (392 with PD) from both home and clinical settings, covering diverse demographics (53.2% female). We used deep learning embeddings from Wav2Vec 2.0, WavLM, and Ima...

Full description

Saved in:
Bibliographic Details
Main Authors: Tariq Adnan, Abdelrahman Abdelkader, Zipei Liu, Ekram Hossain, Sooyong Park, Md Saiful Islam, Ehsan Hoque
Format: Article
Language:English
Published: Nature Portfolio 2025-06-01
Series:npj Parkinson's Disease
Online Access:https://doi.org/10.1038/s41531-025-00956-7
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850114509976371200
author Tariq Adnan
Abdelrahman Abdelkader
Zipei Liu
Ekram Hossain
Sooyong Park
Md Saiful Islam
Ehsan Hoque
author_facet Tariq Adnan
Abdelrahman Abdelkader
Zipei Liu
Ekram Hossain
Sooyong Park
Md Saiful Islam
Ehsan Hoque
author_sort Tariq Adnan
collection DOAJ
description We introduce a framework for screening Parkinson’s disease (PD) using English pangram utterances. Our dataset includes 1306 participants (392 with PD) from both home and clinical settings, covering diverse demographics (53.2% female). We used deep learning embeddings from Wav2Vec 2.0, WavLM, and ImageBind to capture speech dynamics indicative of PD. Our novel fusion model for PD classification aligns different speech embeddings into a cohesive feature space, outperforming baseline alternatives. In a stratified randomized split, the model achieved an AUROC of 88.9% and an accuracy of 85.7%. Statistical bias analysis showed equitable performance across sex, ethnicity, and age subgroups, with robustness across various disease durations and PD stages. Detailed error analysis revealed higher misclassification rates in specific age ranges for males and females, aligning with clinical insights. External testing yielded AUROCs of 82.1% and 78.4% on two clinical datasets, and an AUROC of 77.4% on an unseen general spontaneous English speech dataset, demonstrating versatility in natural speech analysis and potential for global accessibility and health equity.
format Article
id doaj-art-bd543c9a18914c8fb34a028a000910ff
institution OA Journals
issn 2373-8057
language English
publishDate 2025-06-01
publisher Nature Portfolio
record_format Article
series npj Parkinson's Disease
spelling doaj-art-bd543c9a18914c8fb34a028a000910ff2025-08-20T02:36:50ZengNature Portfolionpj Parkinson's Disease2373-80572025-06-0111111810.1038/s41531-025-00956-7A novel fusion architecture for detecting Parkinson’s Disease using semi-supervised speech embeddingsTariq Adnan0Abdelrahman Abdelkader1Zipei Liu2Ekram Hossain3Sooyong Park4Md Saiful Islam5Ehsan Hoque6Department of Computer Science, University of RochesterDepartment of Computer Science, University of RochesterDepartment of Computer Science, University of RochesterDepartment of Computer Science, University of RochesterDepartment of Computer Science, University of RochesterDepartment of Computer Science, University of RochesterDepartment of Computer Science, University of RochesterWe introduce a framework for screening Parkinson’s disease (PD) using English pangram utterances. Our dataset includes 1306 participants (392 with PD) from both home and clinical settings, covering diverse demographics (53.2% female). We used deep learning embeddings from Wav2Vec 2.0, WavLM, and ImageBind to capture speech dynamics indicative of PD. Our novel fusion model for PD classification aligns different speech embeddings into a cohesive feature space, outperforming baseline alternatives. In a stratified randomized split, the model achieved an AUROC of 88.9% and an accuracy of 85.7%. Statistical bias analysis showed equitable performance across sex, ethnicity, and age subgroups, with robustness across various disease durations and PD stages. Detailed error analysis revealed higher misclassification rates in specific age ranges for males and females, aligning with clinical insights. External testing yielded AUROCs of 82.1% and 78.4% on two clinical datasets, and an AUROC of 77.4% on an unseen general spontaneous English speech dataset, demonstrating versatility in natural speech analysis and potential for global accessibility and health equity.https://doi.org/10.1038/s41531-025-00956-7
spellingShingle Tariq Adnan
Abdelrahman Abdelkader
Zipei Liu
Ekram Hossain
Sooyong Park
Md Saiful Islam
Ehsan Hoque
A novel fusion architecture for detecting Parkinson’s Disease using semi-supervised speech embeddings
npj Parkinson's Disease
title A novel fusion architecture for detecting Parkinson’s Disease using semi-supervised speech embeddings
title_full A novel fusion architecture for detecting Parkinson’s Disease using semi-supervised speech embeddings
title_fullStr A novel fusion architecture for detecting Parkinson’s Disease using semi-supervised speech embeddings
title_full_unstemmed A novel fusion architecture for detecting Parkinson’s Disease using semi-supervised speech embeddings
title_short A novel fusion architecture for detecting Parkinson’s Disease using semi-supervised speech embeddings
title_sort novel fusion architecture for detecting parkinson s disease using semi supervised speech embeddings
url https://doi.org/10.1038/s41531-025-00956-7
work_keys_str_mv AT tariqadnan anovelfusionarchitecturefordetectingparkinsonsdiseaseusingsemisupervisedspeechembeddings
AT abdelrahmanabdelkader anovelfusionarchitecturefordetectingparkinsonsdiseaseusingsemisupervisedspeechembeddings
AT zipeiliu anovelfusionarchitecturefordetectingparkinsonsdiseaseusingsemisupervisedspeechembeddings
AT ekramhossain anovelfusionarchitecturefordetectingparkinsonsdiseaseusingsemisupervisedspeechembeddings
AT sooyongpark anovelfusionarchitecturefordetectingparkinsonsdiseaseusingsemisupervisedspeechembeddings
AT mdsaifulislam anovelfusionarchitecturefordetectingparkinsonsdiseaseusingsemisupervisedspeechembeddings
AT ehsanhoque anovelfusionarchitecturefordetectingparkinsonsdiseaseusingsemisupervisedspeechembeddings
AT tariqadnan novelfusionarchitecturefordetectingparkinsonsdiseaseusingsemisupervisedspeechembeddings
AT abdelrahmanabdelkader novelfusionarchitecturefordetectingparkinsonsdiseaseusingsemisupervisedspeechembeddings
AT zipeiliu novelfusionarchitecturefordetectingparkinsonsdiseaseusingsemisupervisedspeechembeddings
AT ekramhossain novelfusionarchitecturefordetectingparkinsonsdiseaseusingsemisupervisedspeechembeddings
AT sooyongpark novelfusionarchitecturefordetectingparkinsonsdiseaseusingsemisupervisedspeechembeddings
AT mdsaifulislam novelfusionarchitecturefordetectingparkinsonsdiseaseusingsemisupervisedspeechembeddings
AT ehsanhoque novelfusionarchitecturefordetectingparkinsonsdiseaseusingsemisupervisedspeechembeddings