Using speech acoustics as biomarkers in the early detection of early-onset parkinson’s disease

Early-Onset Parkinson’s Disease (EOPD) is a neurodegenerative disorder caused by the degeneration of dopaminergic neurons in the substantia nigra. Over 500,000 individuals in the U.S. have been diagnosed with Parkinson’s disease, but the actual number, including undiagnosed cases, may reach 1,000,000. Current subjective motor diagnostic methods, like physical examinations, often result in delayed or inaccurate diagnoses, highlighting the need for alternative approaches. This study explores the potential of speech-based machine learning models for early EOPD detection, leveraging acoustic features such as jitter, shimmer, harmonic-to-noise ratio (HNR), amongst other biomarkers.

We hypothesized that EOPD patients’ speech would exhibit significant differences in multiple acoustic features due to impaired motor control affecting vocal stability. To test this, a Random Forest Classifier was trained on datasets from UCI Parkinson’s and Parkinson’s Telemonitoring, with separate validation and test sets. The model achieved a training accuracy of 94.40% and a test accuracy of 94.81%, outperforming other classifiers such as Support Vector Machines (SVM) and logistic regression. These results confirm the efficacy of using speech features to differentiate EOPD patients from healthy individuals with high precision.

These findings underscore the potential of speech analysis as a simple, cost-effective, non-invasive diagnostic tool for EOPD that can be achieved remotely via telephone or video consultation. Traditional diagnostic methods rely on clinical assessments that may not capture early-stage symptoms, whereas machine learning models trained on speech features provide an objective and accessible alternative. Most existing studies focused on structured speech tasks or patients who have already developed Parkinson’s Disease, which may not have an impact in time to treat EOPD. Future research should incorporate conversational speech, longitudinal data, and EOPD data to improve model robustness and generalizability.

This study establishes a foundation for integrating speech-based biomarkers into clinical screening for EOPD. By refining predictive models and expanding datasets, speech analysis could enhance early detection, reducing misdiagnoses and enabling timely intervention. Additionally, targeted therapies—such as vocal training or neuromodulation—could be developed to address specific speech deficits, improving patient outcomes and quality of life.