Toward Sensor-Based Early Diagnosis of Cognitive Impairment using Poisson Process Models

Area of Science:

• Computational neuroscience
• Gerontology
• Machine learning applications

Background:

• Early diagnosis of cognitive impairment is crucial for timely intervention.
• Traditional assessment methods may not capture subtle, longitudinal changes.
• Home-based sensors offer a non-intrusive method for continuous monitoring.

Purpose of the Study:

• To detect cognitive impairment in elderly adults using home-based sensor data.
• To evaluate the effectiveness of Poisson process (PP) models and machine learning for identifying mild cognitive impairment (MCI) and dementia.
• To compare homogeneous and non-homogeneous PP models for cognitive decline assessment.

Main Methods:

• Analysis of longitudinal time-series sensor data from home environments.
• Application of homogeneous and non-homogeneous Poisson process models.
• Utilizing supervised machine learning algorithms for classification.
• Feature extraction based on sensor signal patterns and task-specific rates.

Main Results:

• The proposed sensor-based approach effectively distinguishes individuals with dementia and MCI from healthy controls.
• Non-homogeneous PP models demonstrated superior performance compared to homogeneous PP models.
• Sensor-based assessment using non-homogeneous PP outperformed expert-based assessment in detecting cognitive impairment.

Conclusions:

• Sensor-based assessment combined with machine learning, particularly non-homogeneous PP models, offers a powerful tool for early cognitive impairment detection in elderly adults.
• This computational method can contribute to the development of ambulatory clinical biomarkers for dementia.
• Findings support the advancement of aging-in-place technologies by enabling proactive cognitive health monitoring.