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

Area of Science:

• Neuroscience
• Quantum Machine Learning

Background:

• Parkinson's disease is the fastest-growing neurodegenerative disorder, with a 50% case increase in two years.
• Early diagnosis is critical to mitigate disease progression impacting speech, memory, and motor functions.
• Existing machine learning models for Parkinson's detection often use single features, leading to errors due to symptom variability.

Purpose of the Study:

• To develop an accurate and accessible method for early Parkinson's disease detection.
• To leverage multi-modal biomarker data for improved classification accuracy.
• To explore the application of quantum machine learning for neurodegenerative disease diagnosis.

Main Methods:

• Utilized the mPower dataset (150,000 samples) including voice, gait, tapping, and demographic data.
• Extracted 64 features and employed a Random Forest model for feature selection (above 80th percentile).
• Designed and implemented a simulatable quantum support vector machine (qSVM) for classification.

Main Results:

• The qSVM model achieved 90% accuracy, a 0.90 F-1 score, and an AUC of 0.98.
• Performance surpassed benchmark models, demonstrating the efficacy of the multi-feature approach.
• The simulatable qSVM architecture runs on standard hardware, enhancing accessibility.

Conclusions:

• The developed qSVM model offers a highly accurate and accessible pathway for global Parkinson's disease screening.
• Multi-modal biomarker analysis combined with quantum machine learning significantly improves diagnostic capabilities.
• This approach addresses limitations of single-feature models and facilitates early intervention.