Sim-to-Real Domain Adaptation for Early Alzheimer's Detection from Handwriting Kinematics Using Hybrid Deep Learning.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Digital Health

Background:

• Alzheimer's disease (AD) causes progressive cognitive and motor decline, with early detection being a significant challenge.
• Traditional diagnostic methods often miss subtle, preclinical signs of AD.
• Digital health and AI offer novel approaches for identifying non-invasive biomarkers for cognitive impairment.

Purpose of the Study:

• To develop and evaluate an AI-driven framework for the early detection of Alzheimer's disease (AD).
• To utilize handwriting motion data captured by a sensor-integrated Smart Pen as a digital biomarker.
• To compare the performance of various machine learning (ML) and deep learning (DL) models for AD detection.

Main Methods:

• A Smart Pen equipped with an inertial measurement unit (MPU-9250) recorded kinematic and dynamic handwriting/drawing signals.
• Evaluated ML algorithms (Logistic Regression, SVM, RF, kNN) and DL architectures (1D-CNN, LSTM, CNN-BiLSTM).
• Implemented a Sim-to-Real Domain Adaptation strategy to augment limited data with synthetic samples.

Main Results:

• Classical ML models showed moderate diagnostic performance (AUC: 0.62-0.76).
• The proposed hybrid CNN-BiLSTM deep learning model achieved superior results with 0.91 accuracy and 0.96 AUC.
• The framework demonstrated effective AD detection using motion-based digital biomarkers.

Conclusions:

• Motion-based digital biomarkers derived from handwriting analysis show significant potential for automated, non-invasive AD detection.
• The AI framework offers a cost-effective and scalable informatics solution for digital cognitive assessment.
• This approach could facilitate earlier identification and intervention for individuals with Alzheimer's disease.