Early Detection of Low Cognitive Scores from Dual-task Performance Data Using a Spatio-temporal Graph Convolutional...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Gerontology

Background:

• Early detection of cognitive decline is crucial for managing dementia progression.
• Dual-task assessments offer a promising avenue for automatic cognitive evaluation.
• Existing dual-task methods suffer from limited feature extraction and motion information, impacting accuracy.

Purpose of the Study:

• To develop an advanced framework for early-stage dementia detection using graph convolutional networks (GCNs).
• To enhance the accuracy and robustness of dual-task based cognitive assessments.
• To address data imbalance issues in cognitive score detection.

Main Methods:

• Utilized graph convolutional networks (GCNs) to extract spatio-temporal features from dual-task performance data.
• Developed a novel loss function optimizing sensitivity and specificity for imbalanced datasets.
• Evaluated the framework on data from 171 subjects across 6 senior living facilities.

Main Results:

• The proposed GCN framework significantly improved the accuracy of detecting low cognitive scores (≤ 23 or ≤ 27).
• The method demonstrated superior performance in both sensitivity and specificity compared to existing approaches.
• The novel loss function effectively handled data imbalance, enhancing detection reliability.

Conclusions:

• The developed graph convolutional network framework offers a robust and accurate method for early dementia detection.
• This approach advances automatic cognitive assessment by leveraging sophisticated spatio-temporal feature extraction.
• Improved early detection capabilities can lead to timely interventions and better management of dementia.