Predicting amyloid status in Primary Progressive Aphasia using explainable artificial intelligence.

Area of Science:

• Computational linguistics
• Artificial intelligence in medicine
• Neurodegenerative disease diagnostics

Background:

• Alzheimer's disease (AD) therapeutics targeting amyloid beta (Aβ) show promise, but Aβ positivity testing is costly.
• Accessible methods are needed to identify patients for timely diagnosis and treatment.

Purpose of the Study:

• To develop and validate a machine learning model for predicting Aβ positivity using connected speech.
• To utilize explainable AI (XAI) to identify linguistic features indicative of Aβ positivity for clinical application.

Main Methods:

• A pre-trained language model (Distil-RoBERTa) was trained on speech samples from 71 patients with Primary Progressive Aphasia (PPA).
• Amyloid positivity was confirmed via cerebrospinal fluid, amyloid PET, or autopsy.
• The LIME algorithm was used for XAI to interpret model predictions and identify linguistic features.

Main Results:

• The Distil-RoBERTa model achieved a mean accuracy of 92% in predicting Aβ positivity.
• XAI methods achieved 97% accuracy and identified novel speech patterns associated with Aβ positivity.
• This represents a 10% improvement over previous research.

Conclusions:

• Connected speech analysis is a valuable, accessible tool for predicting Aβ positivity in patients with PPA.
• XAI successfully revealed novel linguistic biomarkers for potential clinical use.
• Computational linguistic analysis of speech shows significant potential for diagnosing AD and related disorders.