Modeling Users' Cognitive Performance Using Digital Pen Features.

Area of Science:

• Digital pen technology
• Machine learning applications
• Neurocognitive assessment

Background:

• Digital pens capture sketch characteristics and user behavior for machine learning (ML).
• Neurocognitive assessments, like dementia screening, traditionally use pen and paper.
• Real-time analysis of pen input offers objective scoring and aids physicians.

Purpose of the Study:

• To implement and evaluate a state-of-the-art set of over 170 digital pen features.
• To assess the accuracy of ML models in automatically scoring cognitive tests.
• To explore automated ML (AutoML) for optimizing classification performance.

Main Methods:

• Recorded sketch data from 40 subjects using a digital pen for cognitive tests.
• Evaluated 10 ML classifiers (e.g., SVMs, Deep Learning) on the recorded sketch dataset.
• Utilized an automated ML approach for model fine-tuning and performance optimization.

Main Results:

• The digital pen feature set outperformed previous approaches in scoring cognitive tests.
• ML models achieved up to 87.5% accuracy in binary classification tasks for cognitive tests.
• Automated ML demonstrated superior recognition accuracies on the dataset.

Conclusions:

• The developed digital pen feature set and ML models provide accurate, automated scoring of neurocognitive assessments.
• This technology can enhance telemedicine and support physicians by reducing manual scoring.
• The publicly available feature set and findings advance ML applications in medical diagnostics.