Abstract: Objective To describe the eye movement pattern during visual search process,so as to explore whether visual search patterns could be potentially useful to identify children with autism spectrum disorders (ASD). Methods Totally 15 ASD children and 17 typically developing (TD) children matched with intelligence quotient (IQ),gender and chronological age were enrolled in this study from March to July 2017,and their eye-movement data when performing visual search task were recorded.The machine learning method was used to establish a model discriminating ASD from TD,including the fixations,fixation duration and fixation path during the visual search task.The performance of the classification model was assessed by the accuracy,sensitivity and specificity of classifying ASD. Results 1)ASD children showed shorter reaction time in visual search task compared to TD in the condition of target-off(F=3.76,P<0.05).2) By using eye movement patterns,ASD children could be discriminated from TD children accurately,with the classification accuracy of 78.13%,specificity of 70.59% and sensitivity of 86.67%.3) The score of the discriminative features was negatively related to scores of the Childhood Autism Rating Scale (r=-0.497,P=0.030). Conclusion The visual search pattern detected by eye movement could be potentially useful to identify children with ASD.

Key words: visual search, autism spectrum disorders, eye-movement, machine learning

摘要： 目的 分析视觉搜索过程中的眼动模式,探究视觉搜索模式能否作为识别孤独症谱系障碍(ASD)的参考指标。方法 2017年3-7月记录15名ASD儿童和17名性别、年龄、智力匹配的正常对照组儿童在执行视觉搜索任务时的眼动数据,利用机器学习的方法建立区分ASD儿童和正常发育(TD)儿童的分类模型(将被试完成视觉搜索任务时的注视点、注视时间及注视轨迹等眼动指标作为分类器的特征),并评估分类模型的区分准确率、特异性和敏感性。结果 1)当目标物不存在时,ASD儿童完成视觉搜索任务的反应时显著短于TD儿童(F=3.76,P<0.05);2)将视觉搜索的眼动模式作为特征可以有效地对被试类型进行识别,正确率为78.13%,特异性为70.59%,敏感性为86.67%;3)被试在分类器中的特征组合分数与儿童孤独症评定量表(CARS)总分呈现显著负相关 (r=-0.497,P=0.030)。结论 视觉搜索过程中的眼动模式可作为未来ASD诊断的一个重要参考依据。

关键词: 视觉搜索, 孤独症谱系障碍, 眼动, 机器学习