Language biomarker screening using AI: a transdiagnostic approach to the brain

Area of Science:

• Neurolinguistics
• Computational Linguistics
• Artificial Intelligence in Medicine

Background:

• Neurological conditions like dementia, mild cognitive impairment (MCI), left-hemisphere damage (LHD), right-hemisphere damage (RHD), and traumatic brain injury (TBI) present overlapping communication and social interaction deficits.
• Objective and scalable methods for automatically differentiating these conditions using language are lacking.
• Speech analysis offers a non-invasive approach to assess neurological health.

Purpose of the Study:

• To develop comprehensive neurolinguistic measures for differentiating neurological conditions.
• To create a machine learning multiclass screening and language assessment model (NeuroScreen).
• To establish a large comparative database of linguistic biomarkers for future research.

Main Methods:

• Utilized a large database of 291 linguistic biomarkers from speech samples of 1,394 participants (including patients with LHD, RHD, dementia, MCI, TBI, and healthy controls).
• Employed natural language processing (NLP) via the Open Brain AI platform to extract linguistic features (readability, lexical richness, phonology, morphology, syntax, semantics).
• Developed a Deep Neural Network (DNN) for classification and used linear mixed-effects models to identify biomarkers.

Main Results:

• The DNN model achieved high accuracy (up to 91%) in classifying neurological conditions based on linguistic features.
• Distinct neurolinguistic properties were identified: LHD and TBI showed syntax/phonology deficits; MCI exhibited simplification; dementia presented lexico-semantic impairments; RHD had the most preserved profile.
• Quantitative neurolinguistic markers were revealed for each condition.

Conclusions:

• The study provides an automatic detection and classification model (NeuroScreen) for key neurological conditions affecting language.
• Novel, validated neurological markers facilitate differential diagnosis, remote monitoring, and personalized neurological care.
• This approach offers a scalable and objective method for assessing neurological health through language.