Predicting stuttering subtypes from functional connectivity using machine learning

PROJECT SUMMARY Persistent developmental stuttering (PDS) is a neurodevelopmental motor speech disorder that disrupts a speaker’s ability to produce fluent speech, characterized by acoustic disturbances, physical concomitants, and/or avoidance behaviors. Over 50 million adults worldwide present with PDS with negative social, emotional, and professional consequences. Adults who stutter (AWS) report mixed results with speech-therapy intervention, with many who relapse in symptoms. There is high intra-individual and inter-individual variability in stuttering, ranging from different disfluency types, frequency, contexts, and internal and external factors. Although there have been many neuroimaging studies on stuttering, there is still a lack of consensus on the neural mechanisms underlying the disorder. Specifically, there are a limited number of studies that differentiate disfluency distinct stuttering behaviors, which may help characterize the variability in behavioral presentations. Thus, the primary objective of the current project is to investigate the neural networks involved in stuttering through two specific aims: 1) identify neural correlates of distinct disfluency types (repetitions, blocks, prolongations) using resting-state functional magnetic resonance imaging (rs-fMRI) data, and 2) utilize machine learning (ML) and traditional statistical methods to predict predominant disfluency types from resting state functional connectivity. Aim 1 will develop stuttering profiles from speech samples and investigate neural correlates of disfluency types in AWS using resting state functional connectivity data. Aim 2 will predict predominant disfluency types from functional connectivity using ML and traditional statistical methods. The long-term goal is to develop comprehensive stuttering profiles that capture the inherent heterogeneity of AWS for efficacious personalized treatment interventions. The novel application of utilizing ML to predict stuttering disfluency types holds substantial potential to advance the field of stuttering research and clinical practice by identifying patterns from large datasets. This study will further our understanding of the neural substrates of stuttering, which is directly in line with NIDCD’s mission to advance our understanding of normal and disordered speech and improve the lives of individuals with communication disorders. Additionally, the second aim aligns closely with one of NIDCD’s goals to develop ML algorithms to provide novel insights and applications for clinical populations.