Early Parkinson's Disease Detection via Touchscreen Typing Analysis using Convolutional Neural Networks.

Area of Science:

• Neurodegenerative Disorders
• Digital Health
• Machine Learning in Medicine

Background:

• Parkinson's Disease (PD) is a prevalent neurodegenerative disorder with early-stage symptoms often overlooked, delaying diagnosis.
• Remote screening for early motor signs of PD using consumer technology interaction data is an emerging area.
• Quantifying fine-motor skill decline is crucial for early PD detection.

Purpose of the Study:

• To develop and validate a smartphone-based method for early Parkinson's Disease detection.
• To quantify fine-motor skill decline in early PD patients through keystroke dynamics.
• To assess the potential of using Convolutional Neural Networks (CNNs) for remote PD screening.

Main Methods:

• A smartphone-based system was developed to analyze finger interaction with the screen, specifically keystroke typing dynamics.
• Convolutional Neural Networks (CNNs) were employed to analyze typing data for PD detection.
• The method was evaluated on both in-clinic data and real-world, self-reported daily smartphone usage data.

Main Results:

• The system achieved 0.89 Area Under the Curve (AUC) with 0.79 sensitivity/specificity on in-clinic data from early PD patients and controls.
• On a separate "in-the-wild" dataset, the approach demonstrated a generalization ability with 0.79 AUC and 0.74/0.78 sensitivity/specificity.
• Keystroke dynamics from natural typing activity were effectively processed to detect PD.

Conclusions:

• The proposed smartphone-based approach shows significant potential for the remote screening of early Parkinson's Disease motor symptoms.
• This method contributes to the development of digital tools for accessible and timely pathological symptom detection.
• Analyzing natural typing behavior offers a non-invasive pathway for early PD identification and management.