Using Computational Models to Detect Autistic Tendencies for Children from their Story Book Narratives

作     者：Sun, Ruihan Wong, Jasin Chen, Eva E. Chen, Arbee L. P. 

作者机构：Department of Computer Science National Tsing Hua University Hsinchu Taiwan Department of Special Education National Tsing Hua University Hsinchu Taiwan College of Education National Tsing Hua University Hsinchu Taiwan Department of Computer Science and Information Engineering Asia University Taichung Japan 

出 版 物：《Multimedia Tools and Applications》 (Multimedia Tools Appl)

年 卷 期：2025年第84卷第6期

页      面：3175-3196页

核心收录：

学科分类：08[工学] 0835[工学-软件工程] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：We extend our heartfelt gratitude to the children and their families who took part in this study  generously contributing to our dataset. We are also grateful to the Zhulian Elementary School in East District of Hsinchu City and the Hsinchu Autism Association for their cooperation in recruiting the participants for the study and providing the experiment venue.This research was funded by [109-2221-E-468-014-MY3] (awarded to Dr. Arbee L.P. Chen) from National Science and Technology Council  Republic of China 

主　　题：Deep neural networks 

摘      要：Diagnosing autism spectrum disorder (ASD) conventionally demands significant time and resources. Language deficits are key markers of ASD, particularly in constructing narratives. This study leverages computational models to analyze story book narratives from seven children with ASD and 16 typically-developing (TD) peers. By transcribing and training models on limited data using augmentation techniques, our best model achieved over 90% accuracy, sensitivity, and specificity-outperforming previous models by 20% in ASD detection. This research showcases the efficacy of our approach in efficiently assessing language abilities and identifying ASD tendencies. The method holds promise for enhancing diagnostic efficiency and providing comprehensive language evaluations to support children with ASD and their caregivers. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2025.