An analysis of eye-movements during reading for the detection of mild cognitive impairment

Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 1016–1026 Copenhagen, Denmark, September 7–11, 2017. c⃝2017 Association for Computational Linguistics An analysis of eye-movements during reading for the detection of mild cognitive impairment Kathleen C. Fraser1, Kristina Lundholm Fors1, Dimitrios Kokkinakis1, Arto Nordlund2 1The Swedish Language Bank, Department of Swedish 2Department of Psychiatry and Neurochemistry University of Gothenburg, Sweden {kathleen.fraser, kristina.lundholm, dimitrios.kokkinakis, arto.nordlund}@gu.se Abstract We present a machine learning analysis of eye-tracking data for the detection of mild cognitive impairment, a decline in cogni- tive abilities that is associated with an in- creased risk of developing dementia. We compare two experimental configurations (reading aloud versus reading silently), as well as two methods of combining infor- mation from the two trials (concatenation and merging). Additionally, we anno- tate the words being read with information about their frequency and syntactic cate- gory, and use these annotations to generate new features. Ultimately, we are able to distinguish between participants with and without cognitive impairment with up to 86% accuracy. 1 Introduction As the global population ages, the prevalence of dementia is increasing (Prince et al., 2013). The term “dementia” refers to an atypical and patho- logical decline in cognitive abilites, encompassing a range of possible underlying causes. Detecting the onset of dementia as early as possible is impor- tant for a number of reasons, including timely ac- cess to medication and treatment, increasing sup- port for activities of daily living (such as main- taining proper nutrition and hygiene), reducing the individual’s engagement in potentially risky activ- ities (e.g. driving an automobile), and giving in- dividuals, families, and caregivers time to prepare (Solomon and Murphy, 2005; Ashford et al., 2006; Calz`a et al., 2015). In this study, we investigate the possibility of using eye-tracking data and machine learning to detect early, subtle signs of cognitive impairment. Previous work has suggested that changes in eye movements while reading do occur in Alzheimer’s disease (Lueck et al., 2000; Fern´andez et al., 2013; Pereira et al., 2014; Biondi et al., 2017). However, our participants do not have a dementia diagnosis; rather, they have been diagnosed with “mild cog- nitive impairment”, meaning they are starting to show very early signs of cognitive decline, and are at an increased risk of developing dementia. We test the relative merits of collecting eye-tracking data while reading silently and aloud, and explore the idea of augmenting eye-tracking features with linguistic information. We begin by presenting some background infor- mation on cognitive and linguistic changes in de- mentia, and discuss previous work on eye-tracking and natural language processing approaches to de- tecting cognitive decline. We then explain our ex- perimental set-up, feature extraction, and machine learning pipeline. We present results for reading silently and reading aloud, and discuss the over- all implications and interpretation of our results. Finally we acknowledge the limitations of the cur- rent work and suggest areas of future research. 2 Background There are several different types of dementia, Alzheimer’s disease (AD) being the most com- …

Content truncated to 500 words