Detection of cognitive impairment from eSAGE cognitive data using machine learning

Abstract

AbstractBackgroundAlzheimer’s disease is a neurodegenerative disease that affects many people worldwide. Many cognitive screening tests are available, with one example being eSAGE – a self‐administered exam (Scharre et al., 2017)1. eSAGE is a valid and reliable cognitive assessment tool to detect mild cognitive impairment or early dementia. It has been implemented as an app in BrainTest®, with sample screenshots in Figure 1. While eSAGE provides diagnostic results of its own, there is a potential to apply machine learning methods to the responses to obtain better diagnostic results.MethodeSAGE results1 and behavioral data from BrainTest® were obtained for 69 subjects, each with a diagnosis of normal cognition, mild cognitive impairment, or dementia. 84 metrics were extracted, including the time that subjects spent on each test page, and drawing speed and line straightness from the construction test problems (Table 1). The hypothesis was that cognitive impaired subjects would show different behavioral patterns (e.g., spend more time on questions, change answers often) compared to those of normal cognition. Different machine learning methods including logistic regression and gradient boosting are built over the metrics to classify between cognitive impairment and normal cognition.ResultLogistic regression classifies those with cognitive impairment vs normal cognition with a recall of 100% and ROC of 75.32% and classifies those with MCI vs normal cognition with a recall of 76% and ROC of 79.45% (Table 2). The most important features (those highest weighted metrics used to differentiate the diagnostic classification) for both these classifications were similar (Table 3). Individuals with normal cognition had, on average, longer connected line segments (not lifting their finger off the tablet) when performing the Modified Trails B task (Figure 1). They also spent less time completing their verbal fluency and picture naming tasks (Table 3). See Table 2 and 3 for other findings.ConclusionBehavioral features extracted from self‐administered eSAGE with the BrainTest® app are predictive of cognitive impairment diagnosis by logistic regression and gradient boosting. Important features identified suggest executive processing time lengthens as one progresses from normal to impaired cognition.Reference 1Scharre et al. (2017). Alzheimer’s Research & Therapy, 9, 44.