Automated Assessment of Cognitive Tests for Detecting Mild Cognitive Impairment

Oregon Health & Science University, Portland OR

Investigators

Linked publications, trials & patents

Abstract

SUMMARY Detecting the first and earliest stages of cognitive decline, conceptualized as ?mild cognitive impairment? (MCI), has become increasingly important in recent years, as research focuses on delaying the manifestations of more severe stages of decline. It is now well recognized that changes in performance on cognitive tests begin up to ten years (or longer) before clinically apparent symptoms of dementia or functional impairment appear. Conventional assessment methods for assessing cognition Conventional assessment methods typically use the spoken response modality in which subjects are asked to verbally respond to prompts, and examiners carefully listen to these responses and apply test manuals to compute scores. These assessment methods require trained specialists and can be burdensome because they need to performed in clinics, especially so with frequent reassessment. Automating the scoring of assessment is important, not only for alleviating this burden but also for enabling large-scale studies on new intervention methods. Our research goal is to automate detection of MCI by developing, applying, and evaluating a system for automation of verbal (i.e., using spoken responses) cognitive-test scoring. Focusing on four verbal cognitive tests, we develop an extensible system comprising ASR and machine-learning algorithms to automatically score responses, and measure the efficacy of our proposed method by validating the accuracy of the automatically obtained scores against gold-standard, clinically obtained scores (Aim 1). Next, we develop classification methods for detecting individuals with MCI based on all information available in the verbal responses, and validating this classification against gold-standard clinical consensus.