Classification of Children With Developmental Language Disorder Using Task fNIRS Data and Convolutional Neural Network

Abstract

Developmental language disorder (DLD) presents significant clinical challenges and has lasting impacts on children. This study aims to develop a classification model for young children with DLD based on their brain function signals. Children aged 3.0 to 7.0 years participated in this study, including 21 children with DLD and 43 controls. All participants completed functional near-infrared spectroscopy (fNIRS) tasks designed to assess word expression ability (report task) and word comprehension ability (point task). General linear model (GLM) analysis was conducted to compare activation differences across fNIRS channels between the two groups. For DLD classification, a one-dimensional Convolutional Neural Network (CNN) was applied to hemoglobin oxygenation (HbO) signals from three regions of interest (ROIs), which included the bilateral inferior frontal gyrus (encompassing Broca's area), the bilateral temporo-parietal junction (encompassing Wernicke's area), and the bilateral motor cortex. Using HbO signal features the bilateral inferior frontal gyrus during the word expression task, the CNN model achieved a validation F1 score of 72.89%. Similarly, using HbO signal features from from the bilateral temporo-parietal junction during the word comprehension task, the CNN model achieved a validation F1 score of 71.81%. Additionally, children with DLD showed atypical activation in the right temporo-parietal junction area and left inferior frontal gyrus during both tasks. Our findings demonstrate that brain signals recorded during language tasks can effectively differentiate young children with DLD, highlighting the potential of task-based fNIRS as a valuable adjunct in the clinical diagnosis of DLD.