Using Generative Artificial Intelligence to Classify Primary Progressive Aphasia from Connected Speech

Using explainable artificial intelligence to identify linguistic biomarkers of amyloid pathology in primary progressive aphasia

Introduction

Recent success has been achieved in Alzheimer's disease (AD) clinical trials targeting amyloid beta (β), demonstrating a reduction in the rate of cognitive decline. However, testing methods for amyloid-β positivity are currently costly or invasive, motivating the development of accessible screening approaches to steer patients toward appropriate diagnostic tests. Here, we employ a pre-trained language model (Distil-RoBERTa) to identify amyloid-β positivity from a short, connected speech sample. We further use explainable AI (XAI) methods to extract interpretable linguistic features that can be employed in clinical practice.

Methods

We obtained language samples from 74 patients with primary progressive aphasia (PPA) across its three variants. Amyloid-β positivity was established through the analysis of cerebrospinal fluid, amyloid PET, or autopsy. 51% of the sample was amyloid-positive. We trained Distil-RoBERTa for 16 epochs with a batch size of 6 and a learning rate of 5e-5, and used the LIME algorithm to train interpretation models to interpret the trained classifier's inference conditions.

Results

Over ten runs of 10-fold cross-validation, the classifier achieved a mean accuracy of 92%, SD = 0.01. Interpretation models were able to capture the classifier's behavior well, achieving an accuracy of 97% against classifier predictions, and uncovering several novel speech patterns that may characterize amyloid-β positivity.

Discussion

Our work improves previous research which indicates connected speech is a useful diagnostic input for prediction of the presence of amyloid-β in patients with PPA. Further, we leverage XAI techniques to reveal novel linguistic features that can be tested in clinical practice in the appropriate subspecialty setting. Computational linguistic analysis of connected speech shows great promise as a novel assessment method in patients with AD and related disorders.

Impaired language in Alzheimer's disease: A comparison between English and Persian implicates content-word frequency rather than the noun-verb distinction

This study challenges the conventional psycholinguistic view that the distinction between nouns and verbs is pivotal in understanding language impairments in neurological disorders. Traditional views link frontal brain region damage with verb processing deficits and posterior temporoparietal damage with noun difficulties. However, this perspective is contested by findings from patients with Alzheimer's disease (pwAD), who show impairments in both word classes despite their typical temporoparietal atrophy. Notably, pwAD tend to use semantically lighter verbs in their speech than healthy individuals. By examining English-speaking pwAD and comparing them with Persian-speaking pwAD, this research aims to demonstrate that language impairments in Alzheimer's disease (AD) stem from the distributional properties of words within a language rather than distinct neural processing networks for nouns and verbs. We propose that the primary deficit in AD language production is an overreliance on high-frequency words. English has a set of particularly high-frequency verbs that surpass most nouns in usage frequency. Since pwAD tend to use high-frequency words, the byproduct of this word distribution in the English language would be an over-usage of high-frequency verbs. In contrast, Persian features complex verbs with an overall distribution lacking extremely high-frequency verbs like those found in English. As a result, we hypothesize that Persian-speaking pwAD would not have a bias toward the overuse of high-frequency verbs. We analyzed language samples from 95 English-speaking pwAD and 91 healthy controls, along with 27 Persian-speaking pwAD and 27 healthy controls. Employing uniform automated natural language processing methods, we measured the usage rates of nouns, verbs, and word frequencies across both cohorts. Our findings showed that English-speaking pwAD use higher-frequency verbs than healthy individuals, a pattern not mirrored by Persian-speaking pwAD. Crucially, we found a significant interaction between the frequencies of verbs used by English and Persian speakers with and without AD. Moreover, regression models that treated noun and verb frequencies as separate predictors did not outperform models that considered overall word frequency alone in classifying AD. In conclusion, this study suggests that language abnormalities among English-speaking pwAD reflect the unique distributional properties of words in English rather than a universal noun-verb class distinction. Beyond offering a new understanding of language abnormalities in AD, the study highlights the critical need for further investigation across diverse languages to deepen our insight into the mechanisms of language impairments in neurological disorders.

Language Abnormalities in Alzheimer's Disease Arise from Reduced Informativeness: A Cross-Linguistic Study in English and Persian

INTRODUCTION

This research investigates the psycholinguistic origins of language impairments in Alzheimer's Disease (AD), questioning if these impairments result from language-specific structural disruptions or from a universal deficit in generating meaningful content.

METHODS

Cross-linguistic analysis was conducted on language samples from 184 English and 52 Persian speakers, comprising both AD patients and healthy controls, to extract various language features. Furthermore, we introduced a machine learning-based metric, Language Informativeness Index (LII), to quantify informativeness.

RESULTS

Indicators of AD in English were found to be highly predictive of AD in Persian, with a 92.3% classification accuracy. Additionally, we found robust correlations between the typical linguistic abnormalities of AD and language emptiness (low LII) across both languages.

DISCUSSION

Findings suggest AD linguistics impairments are attributed to a core universal difficulty in generating informative messages. Our approach underscores the importance of incorporating biocultural diversity into research, fostering the development of inclusive diagnostic tools.