Answers are grounded in this paper and cite their sources.
Introduction Parkinson's disease is a progressive neurodegenerative disorder that impairs motor and vocal functions. The complementary assessment of these manifestations may provide useful information for non-invasive disease prediction and clinical decision support. This study presents a dual-pipeline stacked ensemble machine learning framework that independently evaluates voice and motor-drawing characteristics for Parkinson's disease prediction. Methods In the voice pipeline, acoustic descriptors of a publicly available speech dataset were subjected to block-wise mutual information and stability selection method, which resulted in 299 physiologically interpretable features. A stacked ensemble model combining histogram gradient boosting and logistic regression was able to attain a test accuracy of 0.96 as well as an area under the receiver operating characteristic curve of 0.987, with mel-frequency cepstral coefficients and tunable-Q wavelet transform features being the most significant contributors for classification. For the motor pipeline, the spiral and wave drawing tasks were encoded through a truncated DenseNet-121 model to get 128-dimensional vectors which represent the tremor patterns, the curvature irregularities and the stroke smoothness. Results The voice pipeline attained a test accuracy of 0.96 area under the curve of 0.987. For the motor pipeline the model achieved an accuracy of 0.84 and an area under the curve of 0.907. Discussion The dual-pipeline strategy reveals that the voice stability modeling and motor-based embeddings offer a non-invasive, interpretable and high-performing framework for early Parkinson's disease prediction.
No takes yet. Share an insight, caveat, or question.
A et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: