Detection of Mild Cognitive Impairment Through Natural Language and Touchscreen Typing Processing

Area of Science:

• Neuroscience
• Digital Health
• Biomarkers

Background:

• Mild cognitive impairment (MCI) is an early stage of Alzheimer's Disease (AD) often missed by current clinical diagnostics.
• Smartphone interaction data offers a non-intrusive method for screening and monitoring MCI progression in daily life.

Purpose of the Study:

• To investigate the diagnostic potential of digital biomarkers derived from fine motor impairment (FMI) and spontaneous written speech (SWS) for detecting MCI.
• To differentiate between individuals with MCI and healthy controls (HC) using unobtrusively collected smartphone data.

Main Methods:

• Analysis of keystroke dynamics during typing using Convolutional Neural Networks (CNNs) to assess FMI.
• Natural Language Processing (NLP) techniques applied to spontaneous written speech to extract linguistic features.
• A cascaded-classifier system combining FMI and NLP features, validated using Leave-One-Subject-Out cross-validation.

Main Results:

• Keystroke dynamics achieved an Area Under Curve (AUC) of 0.78 (SP/SE: 0.64/0.92).
• NLP features achieved an AUC of 0.76 (SP/SE: 0.80/0.71).
• An ensemble model integrating both feature sets yielded an AUC of 0.75 (SP/SE: 0.90/0.60).

Conclusions:

• Digital biomarkers from smartphone interactions show promise for early MCI detection.
• This approach can serve as a highly specific, remote screening tool for cognitive decline in real-world settings.