An Explainable Diagnostic Method for Autism Spectrum Disorder Using Neural Network

• Published in: Journal of Information Science and Engineering Vol. 37; no. 2; pp. 347 - 363
• Main Authors: Zhang, Mingkang, Ma, Yanbiao, Zheng, Linan, Wang, Yuanyuan, Liu, Zhihong, Ma, Jianfeng, Xiang, Qian, Zhang, Kexin, Jiao, Licheng
• Format: Journal Article
• Language: English
• Published: Taipei 社團法人中華民國計算語言學學會 01.03.2021; Institute of Information Science, Academia Sinica
• Subjects: Abnormalities; Artificial neural networks; Autism; Communication skills; Disorders; Magnetic resonance imaging; Medical imaging; Nervous system; Neural networks; Signs and symptoms; magnetic resonance imaging (MRI); network visualization; network interpretation; autism spectrum disorder (ASD); 3D convolutional neural network (3D CNN)
• ISSN: 1016-2364
• DOI: 10.6688/JISE.202103_37(2).0005

Autism spectrum disorder (ASD), also known as autism, is a mental illness caused by disorders of the nervous system. Autism is mainly characterized by developmental disorders, accompanied by abnormalities in social skills, communication skills, interests, and behavioral patterns. Autism cannot be completely cured by existing medical means, and its symptoms can only be relieved through acquired intervention. The best intervention period for autistic patients is before the age of six. But relying on existing methods, most patients with autism have missed the best intervention period when they are diagnosed. In order to allow the subject to be diagnosed with autism in a timely manner, we proposed a method that uses a deep neural network to analyze the subject's magnetic resonance imaging (MRI) and evaluate the performance for early screening of ASD. Our primary analysis of patients with functional magnetic resonance imaging (fMRI) also compared with structural magnetic resonance imaging (sMRI). Experiments have shown that fMRI is more sensitive to autism than sMRI. In addition, we explain the classification results of fMRI.