Language impairment is an important biomarker of neurodegenerative disorders such as Alzheimer’s disease (AD). Artificial intelligence (AI), particularly natural language processing (NLP), has recently been increasingly used for early prediction of AD through speech. Yet, relatively few studies exist on using large language models, especially GPT-3, to aid in the early diagnosis of dementia. In this work, we show for the first time that GPT-3 can be utilized to predict dementia from spontaneous speech. Specifically, we leverage the vast semantic knowledge encoded in the GPT-3 model to generate text embedding, a vector representation of the transcribed text from speech, that captures the semantic meaning of the input. We demonstrate that the text embedding can be reliably used to (1) distinguish individuals with AD from healthy controls, and (2) infer the subject’s cognitive testing score, both solely based on speech data. We further show that text embedding considerably outperforms the conventional acoustic feature-based approach and even performs competitively with prevailing fine-tuned models. Together, our results suggest that GPT-3 based text embedding is a viable approach for AD assessment directly from speech and has the potential to improve early diagnosis of dementia.

There is currently no simple, widely available screening method for Alzheimer’s disease (AD), partly because the diagnosis of AD is complex and typically involves expensive and sometimes invasive tests not commonly available outside highly specialized clinical settings. Here, we developed an artificial intelligence (AI)-powered end-to-end system to detect AD and predict its severity directly from voice recordings. At the core of our system is the pre-trained data2vec model, the first high-performance self-supervised algorithm that works for speech, vision, and text. Our model was internally evaluated on the ADReSSo (Alzheimer’s Dementia Recognition through Spontaneous Speech only) dataset containing voice recordings of subjects describing the Cookie Theft picture, and externally validated on a test dataset from DementiaBank. The AI model can detect AD with average area under the curve (AUC) of 0.846 and 0.835 on held-out and external test set, respectively. The model was well-calibrated (Hosmer-Lemeshow goodness-of-fit p-value = 0.9616). Moreover, the model can reliably predict the subject’s cognitive testing score solely based on raw voice recordings. Our study demonstrates the feasibility of using the AI-powered end-to-end model for early AD diagnosis and severity prediction directly based on voice, showing its potential for screening Alzheimer’s disease in a community setting.

Human prescription drug labeling contains a summary of the essential scientific information needed for the safe and effective use of the drug and includes the Prescribing Information, FDA-approved patient labeling (Medication Guides, Patient Package Inserts and/or Instructions for Use), and/or carton and container labeling. Drug labeling contains critical information about drug products, such as pharmacokinetics and adverse events. Automatic information extraction from drug labels may facilitate finding the adverse reaction of the drugs or finding the interaction of one drug with another drug. Natural language processing (NLP) techniques, especially recently developed Bidirectional Encoder Representations from Transformers (BERT), have exhibited exceptional merits in text-based information extraction. A common paradigm in training BERT is to pretrain the model on large unlabeled generic language corpora, so that the model learns the distribution of the words in the language, and then fine-tune on a downstream task. In this paper, first, we show the uniqueness of language used in drug labels, which therefore cannot be optimally handled by other BERT models. Then, we present the developed PharmBERT, which is a BERT model specifically pretrained on the drug labels (publicly available at Hugging Face). We demonstrate that our model outperforms the vanilla BERT, ClinicalBERT and BioBERT in multiple NLP tasks in the drug label domain. Moreover, how the domain-specific pretraining has contributed to the superior performance of PharmBERT is demonstrated by analyzing different layers of PharmBERT, and more insight into how it understands different linguistic aspects of the data is gained.