An FPGA Prototype for Parkinson’s Disease Detection Using Machine Learning on Voice Signal

Parkinson’s disease (PD) is a chronic neurological disorder caused by a reduction in dopamine levels in the brain. Early diagnosis is crucial for effective treatment. This paper proposes an efficient machine learning model for PD detection using voice-based features, which offer a non-invasive, cost-effective, and accessible alternative to complex imaging methods. To enhance classification performance and reduce computational complexity, we evaluate three feature selection algorithms—Chi-squared (χ<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>), Minimum Redundancy Maximum Relevance (mRMR), and Analysis of Variance (ANOVA) — and adopt an incremental feature selection approach, where each feature set increment is assessed across five classifiers: K-Nearest Neighbors (KNN), Decision Tree (DT), Artificial Neural Network (ANN), Logistic Regression (LR), and Support Vector Machine (SVM). To address dataset imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) is applied. Among the evaluated approaches, the ANOVA algorithm yields the most optimal feature set, with the top 10 features enabling the Quadratic SVM to achieve a classification accuracy of 98.86%. The Quadratic SVM, optimized for minimal power consumption, is implemented on a Nexys A7 FPGA. This setup achieves a dynamic power consumption of just 23 mW and delivers a performance acceleration of 155× compared to equivalent computations on a 6th-generation Intel i5 processor. By combining a high-accuracy machine learning model with an energy-efficient FPGA implementation, our approach offers a powerful and portable solution for real-time PD detection.