Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Deep Learning-Based Detection of Cognitive Impairment from Passive Smartphone Sensing with Routine-Aware Augmentation and Demographic Personalization

Yufei Shen, Ji Hwan Park, Minchao Huang, Jared F. Benge, Justin F. Rousseau, Rosemary A. Lester-Smith, Edison Thomaz

TL;DR

This work tackles the challenge of early cognitive impairment detection amid infrequent clinical assessments by leveraging passive smartphone sensing and an LSTM-based time-series model. It introduces routine-aware augmentation and demographic personalization to improve generalization across participants, achieving a substantial boost in prediction performance under leave-one-participant-out validation. The best approach, fusing sensing features with demographic context, reaches an AUPRC of 0.766 and an AUC of 0.780, outperforming baselines and prior studies. The findings support scalable, continuous cognitive monitoring in aging populations using passively collected behavioral data and personalized training strategies.

Abstract

Early detection of cognitive impairment is critical for timely diagnosis and intervention, yet infrequent clinical assessments often lack the sensitivity and temporal resolution to capture subtle cognitive declines in older adults. Passive smartphone sensing has emerged as a promising approach for naturalistic and continuous cognitive monitoring. Building on this potential, we implemented a Long Short-Term Memory (LSTM) model to detect cognitive impairment from sequences of daily behavioral features, derived from multimodal sensing data collected in an ongoing one-year study of older adults. Our key contributions are two techniques to enhance model generalizability across participants: (1) routine-aware augmentation, which generates synthetic sequences by replacing each day with behaviorally similar alternatives, and (2) demographic personalization, which reweights training samples to emphasize those from individuals demographically similar to the test participant. Evaluated on 6-month data from 36 older adults, these techniques jointly improved the Area Under the Precision-Recall Curve (AUPRC) of the model trained on sensing and demographic features from 0.637 to 0.766, highlighting the potential of scalable monitoring of cognitive impairment in aging populations with passive sensing.

Deep Learning-Based Detection of Cognitive Impairment from Passive Smartphone Sensing with Routine-Aware Augmentation and Demographic Personalization

TL;DR

This work tackles the challenge of early cognitive impairment detection amid infrequent clinical assessments by leveraging passive smartphone sensing and an LSTM-based time-series model. It introduces routine-aware augmentation and demographic personalization to improve generalization across participants, achieving a substantial boost in prediction performance under leave-one-participant-out validation. The best approach, fusing sensing features with demographic context, reaches an AUPRC of 0.766 and an AUC of 0.780, outperforming baselines and prior studies. The findings support scalable, continuous cognitive monitoring in aging populations using passively collected behavioral data and personalized training strategies.

Abstract

Early detection of cognitive impairment is critical for timely diagnosis and intervention, yet infrequent clinical assessments often lack the sensitivity and temporal resolution to capture subtle cognitive declines in older adults. Passive smartphone sensing has emerged as a promising approach for naturalistic and continuous cognitive monitoring. Building on this potential, we implemented a Long Short-Term Memory (LSTM) model to detect cognitive impairment from sequences of daily behavioral features, derived from multimodal sensing data collected in an ongoing one-year study of older adults. Our key contributions are two techniques to enhance model generalizability across participants: (1) routine-aware augmentation, which generates synthetic sequences by replacing each day with behaviorally similar alternatives, and (2) demographic personalization, which reweights training samples to emphasize those from individuals demographically similar to the test participant. Evaluated on 6-month data from 36 older adults, these techniques jointly improved the Area Under the Precision-Recall Curve (AUPRC) of the model trained on sensing and demographic features from 0.637 to 0.766, highlighting the potential of scalable monitoring of cognitive impairment in aging populations with passive sensing.

Paper Structure

This paper contains 28 sections, 1 equation, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Works
  3. Digital Phenotyping for Cognitive Impairment
  4. Deep Learning for Time Series in Health Sensing
  5. Data Augmentation and Model Personalization
  6. Data Acquisition and Processing
  7. Study Protocol
  8. Smartphone Sensing Application
  9. Passive Sensing Features
  10. Activity
  11. Pedometer and Gait
  12. Location
  13. Smartphone and App Usage
  14. Typing
  15. Communication
  16. ...and 13 more sections

Figures (3)

  • Figure 1: Overall architecture of the LSTM model for detecting cognitive impairment from 30-day sequences of daily passive sensing features.
  • Figure 2: t-SNE visualization of participants' daily passive sensing features from days with sufficient sensing coverage, color-coded by participant ID.
  • Figure 3: Scatter plots of age and education for male and female participants, color-coded by cognitive status.