Jiannan Kang1, Xiaoya Han2, Jiajia Song1, Zikang Niu3, Xiaoli Li4. 1. College of Electronic & Information Engineering, Hebei University, Baoding, China. 2. School of Information Science & Engineering, Yanshan University, Qinhuangdao, China. 3. State Key Laboratory of Cognitive Neuroscience and Learning, Beijing Normal University, Beijing, China. 4. State Key Laboratory of Cognitive Neuroscience and Learning, Beijing Normal University, Beijing, China. Electronic address: xiaoli@bnu.edu.cn.
Abstract
OBJECTIVE: To identify autistic children, we used features extracted from two modalities (EEG and eye-tracking) as input to a machine learning approach (SVM). METHODS: A total of 97 children aged from 3 to 6 were enrolled in the present study. After resting-state EEG data recording, the children performed eye-tracking tests individually on own-race and other-race stranger faces stimuli. Power spectrum analysis was used for EEG analysis and areas of interest (AOI) were selected for face gaze analysis of eye-tracking data. The minimum redundancy maximum relevance (MRMR) feature selection method combined with SVM classifiers were used for classification of autistic versus typically developing children. RESULTS: Results showed that classification accuracy from combining two types of data reached a maximum of 85.44%, with AUC = 0.93, when 32 features were selected. LIMITATIONS: The sample consisted of children aged from 3 to 6, and no younger patients were included. CONCLUSIONS: Our machine learning approach, combining EEG and eye-tracking data, may be a useful tool for the identification of children with ASD, and may help for diagnostic processes.
OBJECTIVE: To identify autisticchildren, we used features extracted from two modalities (EEG and eye-tracking) as input to a machine learning approach (SVM). METHODS: A total of 97 children aged from 3 to 6 were enrolled in the present study. After resting-state EEG data recording, the children performed eye-tracking tests individually on own-race and other-race stranger faces stimuli. Power spectrum analysis was used for EEG analysis and areas of interest (AOI) were selected for face gaze analysis of eye-tracking data. The minimum redundancy maximum relevance (MRMR) feature selection method combined with SVM classifiers were used for classification of autistic versus typically developing children. RESULTS: Results showed that classification accuracy from combining two types of data reached a maximum of 85.44%, with AUC = 0.93, when 32 features were selected. LIMITATIONS: The sample consisted of children aged from 3 to 6, and no younger patients were included. CONCLUSIONS: Our machine learning approach, combining EEG and eye-tracking data, may be a useful tool for the identification of children with ASD, and may help for diagnostic processes.
Authors: James C McPartland; Matthew D Lerner; Anjana Bhat; Tessa Clarkson; Allison Jack; Sheida Koohsari; David Matuskey; Goldie A McQuaid; Wan-Chun Su; Dominic A Trevisan Journal: J Autism Dev Disord Date: 2021-05-27