Detection of Cognitive Impairment From eSAGE Cognitive Data Using Machine Learning
Loading...
Date
2022-05
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
The Ohio State University
Abstract
Background: Alzheimer's disease (AD) is a neurodegenerative disease that affects many people worldwide. Early detection of cognitive decline is key to slowing down AD progression. eSAGE, implemented as an app in BrainTest®, is a valid and reliable cognitive screening tool to detect mild cognitive impairment or early dementia. Using data collected in eSAGE, there is a potential to apply machine learning methods to obtain better diagnostic results.
Methods: eSAGE scores and behavioral data from BrainTest® were obtained for 66 subjects, each with a diagnosis of normal cognition, mild cognitive impairment, or dementia. Behavioral features such as the time spent on each test page, drawing speed and average stroke length were extracted for each subject. Scores that the subjects received for each test question were also extracted. The
hypothesis was that cognitively impaired subjects would show different behavioral patterns (e.g., spend more time on questions) compared to those of normal cognition, which could better predict cognitive impairment than eSAGE scores alone. Logistic regression and gradient boosting models were trained using these features to detect cognitive impairment. Performance was evaluated using
five-fold cross validation, with accuracy, precision, recall, F1 score, and AUC score as evaluation metrics.
Results: Logistic regression with feature selection achieved an AUC of 92.88%, a recall of 89.11% and an F1 of 84.93% using both behavioral and scoring features together to classify cognitive impairment vs. normal cognition, demonstrating the strong potential of using these two features together in detecting cognitive impairment. Logistic regression using these two features also achieved an AUC of 88.70% in detecting mild cognitive impairment from normal cognition, and an AUC of 99.20% in detecting dementia from normal cognition. One behavioral feature describing the average stroke length was identified as particularly useful, together with another four scoring features. Using only these five features, logistic regression achieved an even better AUC of 94.06% in detecting cognitive impairment.
Conclusions: Behavioral features and scoring features are predictive of cognitive impairment. They demonstrated great potential to be further integrated into BrainTest® and, with machine learning tools, enable more accurate detection for cognitive impairment.
Description
Keywords
self-administered cognitive test, machine learning, cognitive impairment, alzheimer's disease