ABC-AI
ABC-AI is an interdisciplinary project between the Artificial Intelligence Institute at the University of South Carolina (AIISC@UofSC) and the Aging Brain Cohort at University of South Carolina (ABC@UofSC) developed to identify cognitive decline/impairment among healthy older adults.
Contents
People
Team: Prof. Roger Newman-Norlund, Sara Sayers, Sarah Newman-Norlund, Prof. Amit P. Sheth, Julius Fridriksson
Graduate Students: Usha Lokala
About Our Work
Aims: The purpose of this work is to analyze cognitive decline characteristics of healthy older adults, analyze the language features in discourse transcripts, and identify factors in the speech to text analysis that contributed to the severity of cognitive impairment. These are the problems that Artificial Intelligence is solving in the Aphasia project: 1. Which aspects of spoken language can predict cognitive impairment and/or decline? 2. What is the relationship between cognition, language, and brain health in normal aging? 3. What language factors related to brain health can act as predictors of cognitive decline in healthy older adults? 4. How Natural Language Understanding can help create models of language production that may predict changes in cognition, language, and brain health?
Methods: The team created a public repository that houses multimodal data collected as part of the Aging Brain Cohort the study being conducted at the University of South Carolina (ABC@UofSC). Ultimately, the ABC@UofSC Repository will contain diverse data from cross-sectional (N=800, age=20–80) and longitudinal (N=200, age=60–80, interesting interval=4 years) samples of healthy South Carolinians which include socio-demographic data, raw and preprocessed functional (resting-state and task-based fMRI, ASL) and structural (T1, T2 FLAIR, DWI, SWI) MRI data, raw, and preprocessed resting-state EEG, comprehensive blood work, measures of physical and sensory function, genetic data derived from whole blood and buccal swabs, and results from a unique constellation of social, emotional, cognitive, and language measures. Currently, data has been collected from 65 participants (ages 60–80) in the longitudinal arm of the project. The current manual intensive analysis of discourse transcripts is automated to assist the clinician to make informed decisions. This automatic pipeline (Figure 2) includes extracting language features, and speech features, building regression models to predict MoCA scores, and classification models to predict Cognitive impairment levels (mild, moderate, severe).
Results: The Aphasia study pipeline helps in automating the NLP (Natural Language Processing) feature extraction, Main concept analysis, MoCA score assessment, predicting cognitive impairment, and explainable predictions.
Conclusions: Data from the ABC@UofSC Repository are easily accessible upon request (abc.sc.edu), and our publicly available statistics and visualization tools provide collaborating researchers with the ability to identify associations between brain structure and function in relation to genetic variation and behavioral mea- sures across the adult lifespan with unprecedented ease and rapidity.
Overview
AI plays a key role in modernizing clinical assessment or biomarker identification. Current application areas are autism, aphasia, cognitive decline, ADHD, and mental health. Using AI, we can address the limitations of the current manual methods for assessments through automation (reducing the time it takes to get back to the patients, and the time it takes to collect the data and analyze/score the data) and making the analysis more consistent. This video shows our collaboration with ABC (abc.sc.edu)- a major UofSC project studying age-related cognitive decline.
- Develop Aphasia study pipeline for automated processing of Language measures, Semantic, Syntactic, Lexical, and Speech features of discourse transcripts (Cat Rescue and Cookie Theft) Figure 2.
- Identify language factors related to brain health that can act as predictors of cognitive decline in healthy older adults.
- Create models of language production that predict changes in cognition, language, and brain health.