Recent success has been achieved in Alzheimer’s disease (AD) therapeutics targeting amyloid beta (Aβ), demonstrating a significant reduction in rates of cognitive decline. However, testing methods for Aβ positivity are costly, which motivates the development of accessible approaches to promptly steer patients toward appropriate diagnostic tests. Here, we employ a pre-trained language model (Distil-RoBERTa) to predict Aβ positivity from short, connected speech samples. We further use explainable artificial intelligence (XAI) methods to extract interpretable linguistic features that can be employed in clinical practice. We obtained language samples from 71 patients with Primary Progressive Aphasia across its three variants. Aβ positivity was established through the analysis of cerebrospinal fluid, amyloid PET, or autopsy. 51% of the participants were 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. Over ten runs of 10-fold cross-validation, the classifier achieved a mean accuracy of 92%, SD = 0.01, an absolute improvement of 10% compared to prior research aiming to accomplish similar goals. 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 Aβ positivity. Our work indicates that connected speech is a valuable diagnostic input for predicting the presence of Aβ 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 remarkable promise as a novel assessment method in patients with AD and related disorders.
Robertson et al. (Mon,) studied this question.