A quantum inspired machine learning approach for multimodal Parkinson's disease screening

Abstract

Parkinson's disease, currently the fastest-growing neurodegenerative disorder globally, has seen a 50% increase in cases within just two years. As disease progression impairs speech, memory, and motor functions over time, early diagnosis is crucial for preserving patients' quality of life. Although machine-learning-based detection has shown promise for detecting Parkinson's disease, most studies rely on a single feature for classification and can be error-prone due to the variability of symptoms between patients. To address this limitation we utilized the mPower dataset, which includes 150,000 samples across four key biomarkers: voice, gait, tapping, and demographic data. From these measurements, we extracted 64 features and trained a baseline Random Forest model to select the features above the 80th percentile. For classification, we designed a simulatable quantum support vector machine (qSVM) that detects high-dimensional patterns, leveraging recent advancements in quantum machine learning. With this novel and simulatable architecture that can be run on standard hardware rather than resource-intensive quantum computers, our model achieves an accuracy of 90%, F-1 score of 0.90, and an AUC of 0.98-surpassing benchmark models. Utilizing an innovative classification framework built on a diverse set of features, our model offers a pathway for accessible global Parkinson's screening.

Fig. 1

Feature importance values for features above the 80th percentile in a baseline random forest model. This figure displays the feature importance values for variables ranking above the 80th percentile in a baseline random forest model. The feature importance metric quantifies each variable’s contribution to the model’s ability to predict outcomes. Age emerges as the most significant feature, with a markedly higher importance score compared to others. This indicates its dominant role in the predictive model, potentially serving as a key demographic indicator for the task at hand. Following Age, features related to motor activity, particularly tapping metrics, demonstrate high importance. These include Tap Consistency, Left Taps, Total Taps, and Right Taps, which collectively reflect fine motor coordination and variability in tapping behavior. A notable inclusion is Voice: Spectral Centroid Mean, reflecting the role of vocal biomarkers in the analysis. Figure was generated using version 3.9.2 of the Python package Matplotlib (https://pypi.org/project/matplotlib/).

Fig. 2

Radar chart for proposed model and benchmark comparison across accuracy, precision, recall, F1 score and ROC/AUC score. This figure presents a radar chart comparing the performance of the proposed model and various benchmark machine learning algorithms across five metrics: accuracy, ROC/AUC, precision, recall, and F1 score. The proposed model, represented in red, demonstrates consistently high values across all metrics. In comparison, models such as Linear SVM and Logistic Regression show moderate performance with balanced scores, while Naive Bayes and Random Forest display lower performance levels, as indicated by their proximity to the center of the chart. Neural network-based approaches, including DNN and CNN, exhibit competitive performance, with DNN achieving slightly higher precision and recall. Quantum machine learning models, such as those utilizing the Z Feature Map, perform well in certain metrics like accuracy and F1 score, though their overall scores remain distinct from the proposed model. The radar chart provides a comprehensive visual comparison of the models, highlighting variations in performance across the selected evaluation metrics. Figure was generated using version 3.9.2 of the Python package Matplotlib (https://pypi.org/project/matplotlib/).

Fig. 3

Feature importance in Parkinson’s disease detection model. SHAP values indicate feature impact on model predictions, with positive values (right) increasing disease probability and negative values (left) decreasing it. Color intensity represents relative feature importance.

Fig. 4

Example qubit encoding of one feature of the data. This qubit represents one of 15 qubits in the qSVM kernel circuit, with one qubit corresponding to each feature of the dataset. SVM kernels are designed to output a measure of similarity between two inputs x₁ and x₂. This figure shows the overlap computation between two inputs from the training set, using the gates R_Z(), , R_Z(x₁ - x₂), and R_Z(). These gates represent a quantum mechanical decomposition of R_Y(x₁) followed by R_Y(-x₂). Finally, the measurement gate at the end provides an output corresponding to the similarity between the inputs x₁ and x₂. Figure was generated using version 1.2.4 of the Qiskit Python library (https://pypi.org/project/qiskit/).

Fig. 5

Quantum-Inspired Support Vector Machine (qSVM) Framework. The diagram outlines a qSVM architecture where classical data is processed through a hybrid quantum-classical pipeline: Data Preparation Raw data is cleaned, normalized, and features are extracted for quantum encoding. Quantum Feature Mapping: Classical data is embedded into a quantum feature space using statevector simulation. Quantum Processing: Quantum kernels, measuring data similarity in the quantum space, are computed. Quantum Circuit: Quantum operations on qubits are executed, including the decomposition of the RY gate into RZ and √x gates. SVM Training: A classical optimizer employs the quantum kernel matrix to determine the optimal separating hyperplane. Prediction & Evaluation: New data is classified and performance is evaluated using metrics such as accuracy, precision, and recall. The pipeline follows the sequence: Input Data → Feature Encoding → Quantum Circuit → Kernel Calculation → SVM Optimization → Classification. This approach integrates quantum-enhanced feature representation with traditional SVM for improved classification.