Learning Language and Acoustic Models for Identifying Alzheimer’s Dementia From Speech

Abstract

Alzheimer’s dementia (AD) is a chronic neurodegenerative illness that manifests in a gradual decline of cognitive function. Early identification of AD is essential for managing the ensuing cognitive deficits, which may lead to a better prognostic outcome. Speech data can serve as a window into cognitive functioning and can be used to screen for early signs of AD. This paper describes methods for learning models using speech samples from the DementiaBank database, for identifying which subjects have Alzheimer’s dementia. We consider two machine learning tasks: 1) binary classification to distinguish patients from healthy controls, and 2) regression to estimate each subject’s Mini-Mental State Examination (MMSE) score. To develop models that can use acoustic and/or language features, we explore a variety of dimension reduction techniques, training algorithms, and fusion strategies. Our best performing classification model, using language features with dimension reduction and regularized logistic regression, achieves an accuracy of 85.4% on a held-out test set. On the regression task, a linear regression model trained on a reduced set of language features achieves a root mean square error (RMSE) of 5.62 on the test set. These results demonstrate the promise of using machine learning for detecting cognitive decline from speech in AD patients.

Multimodal Deep Learning Models for Detecting Dementia From Speech and Transcripts

“…Regarding BERT + ViT + Co-Attention, it improves the RMSE scores of all the existing research initiatives, except Bimodal Network (Ensembled Output) (Koo et al, 2020 ), by 0.14–1.41. In terms of the Multimodal BERT - eGeMAPS, Multimodal BERT - ViT, and Multimodal BERT - eGeMAPS + ViT, it seems that these architectures are rather complex for the MMSE regression task improving the RMSE score of only one research work (Shah et al, 2021 ).…”

“…Shah et al ( 2021 ) used also an ensemble method to predict AD patients. Specifically, after training acoustic and language models, they chose the three best performing acoustic models and the best performing language model.…”

Detecting Alzheimer’s Disease Using Interactional and Acoustic Features from Spontaneous Speech

“…Table 3 shows the ranked features from both feature sets, together with their Pearson's correlation (r) with the diagnosis class; due to space limitations, we only show the top 10 features. 2 The most significant acoustic feature was LogHNR, known to be important in acoustic analysis for the diagnosis of pathological voices; loudness, raw fundamental frequency, variation in jitter, intensity, and LogHNR all positively correlate with AD and have been reported as useful features in literature for Dementia [31,9]. Among interactional features, lapses are positively correlated with AD, indicating that patients find trouble continuing topics, resulting in delays with interviewers initiating a new topic.…”

“…Its highest incidence is among adults due to age as a risk factor: one in every six individuals over the age of 80 is likely to develop AD and the number of cases over the age of 60 is doubling every 45 years [1]. Early recognition of cognitive decline could be helpful in managing pre-stage AD thus allowing better quality of life for elderly patients and their caregivers [2].…”

“…Some work has taken a multimodal approach to AD classification: Campbell et al [14] examined two fusion strategies with linguistic features and acoustic features, achieving 75% accuracy. Shah et al [2] used a weighted majority-vote ensemble algorithm for classification and chose the best performing language model with the three best performing acoustic models, giving final prediction accuracy of 83%.…”