Longitudinal voice biomarker trajectory modelling for Parkinson's disease severity: domain-adaptive transfer learning on mPower real-world smartphone data - Takeaways - MDSpire

Longitudinal voice biomarker trajectory modelling for Parkinson's disease severity: domain-adaptive transfer learning on mPower real-world smartphone data

  • By

  • Deepika Roselind Johnson

  • G. Logeswari

  • July 7, 2026

  • 0 min

Share

  • 1

    DAT-PD is a domain-adaptive transformer model designed to predict continuous Parkinson's disease severity from smartphone voice recordings.

  • 2

    The model was developed using the mPower dataset, which includes over 58,000 voice recordings from 5,800 participants.

  • 3

    DAT-PD achieved a mean absolute error of 2.74 MDS-UPDRS units, outperforming six state-of-the-art baseline models.

  • 4

    eGeMAPS features significantly improved model performance compared to MFCC-only representations, reducing mean absolute error.

  • 5

    The model maintains robustness under noisy conditions, supporting unsupervised home-based monitoring of Parkinson's disease.

Original Source(s)

Longitudinal voice biomarker trajectory modelling for Parkinson's disease severity: domain-adaptive transfer learning on mPower real-world smartphone data

  • by Deepika Roselind Johnson, G. Logeswari

  • July 7, 2026

Related Content

Utilizing conditional generative adversarial network to generate head MRA based on nonvascular sequences: comparative study of single-modality and multi-modality methods

Clinical implications of free triiodothyronine levels and diagnostic revisions in antibody-negative autoimmune encephalitis

Collateral circulation after revascularization in moyamoya disease: influencing factors and underlying mechanisms