Analysis of Hierarchical Multi-Content Text Classification Model on B-SHARP Dataset for Early Detection of Alzheimer’s Disease

Proceedings of the 1st Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 10th International Joint Conference on Natural Language Processing, pages 358–365 December 4 - 7, 2020. c⃝2020 Association for Computational Linguistics 358 Analysis of Hierarchical Multi-Content Text Classification Model on B-SHARP Dataset for Early Detection of Alzheimer’s Disease Renxuan A. Li Computer Science Emory University Atlanta, GA, USA albert.li@emory.edu Ihab Hajjar Neurology Emory University Atlanta, GA, USA ihajjar@emory.edu Felicia Goldstein Neurology Emory University Atlanta, GA, USA fgoldst@emory.edu Jinho D. Choi Computer Science Emory University Atlanta, GA, USA jchoi31@emory.edu Abstract This paper presents a new dataset, B-SHARP, that can be used to develop NLP models for the detection of Mild Cognitive Impairment (MCI) known as an early sign of Alzheimer’s disease. Our dataset contains 1-2 min speech segments from 326 human subjects for 3 topics, (1) daily activity, (2) room environment, and (3) picture description, and their transcripts so that a total of 650 speech segments are collected. Given the B-SHARP dataset, several hierarchical text classification models are developed that jointly learn combinatory features across all 3 topics. The best performance of 74.1% is achieved by an ensemble model that adapts 3 types of trans- former encoders. To the best of our knowledge, this is the first work that builds deep learning- based text classification models on multiple contents for the detection of MCI. 1 Introduction Alzheimer’s Disease (AD) is a progressive neurode- generative disorder that is associated with memory loss and declines in major brain functions including semantic and pragmatic levels of language process- ing (Vestal et al., 2006; Ferris and Farlow, 2013). Traditional cognitive assessments such as positron emission tomography or cerebrospinal fluid analy- sis are expensive and time-consuming (Fyffe et al., 2011). This may cause delay in treating AD, known to be irreversible and incurable (Korczyn, 2012), and put an increasing pressure on public health, es- pecially for seniors whose life expectancy is rapidly growing yet are more likely to develop AD. Thus, it is crucial to find a more intelligent way of detecting AD in the earliest stage possible (Karr et al., 2018). Mild Cognitive Impairment (MCI) is considered the first phase that patients start having biomarker evidence of brain changes that can eventually lead to AD (Albert et al., 2011). MCI involves subtle language changes from impairment in reasoning that may not be noticeable to people other than friends and relatives. Because of this, the detec- tion of MCI is a much more challenging task than detecting dementia (Suzman and Beard, 2011). Re- cent studies in NLP have shown that it is possible to detect early stages of AD by analyzing patients’ language patterns; however, most previous works have focused on the detection of dementia instead and researches tackling the detection of MCI have been based on relatively small datasets (Section 2). This paper presents a new dataset that comprises three types of speech segments from both normal controls and MCI patients (Section 3). Then, a hierarchical text classification model …

Content truncated to 500 words