自闭症(autism spectrum disorder, ASD)的早期诊断对自闭症的康复至关重要。近年来,利用机器学习和眼动追踪技术对儿童进行早期自闭症诊断已成为自闭症领域的研究热点。该文在前人工作的基础上,提出了一种基于相对瞳孔响应的眼动数据预处理和瞳孔响应特征提取方法,并使用朴素Bayes算法构建了自闭症分类模型。在现有3~6岁的25名ASD儿童和50名典型发育(typical development, TD)儿童的眼动数据集Autism Detection Dataset上进行了验证,发现了自闭症儿童异常的瞳孔响应。实验结果表明:该文提出的方法在深入挖掘瞳孔特征并仅使用瞳孔特征建模的同时,实现了90.67%的平均分类准确率与92.24%的平均AUC值,优于前人同时使用瞳孔特征、注视行为特征来建模实现的82.2%的平均准确率的结果和同时使用注视行为特征、运动学特征来建模实现的78%的平均准确率的结果,兼具简易和高性能的优点。这不仅证明了该方法的有效性,还提高了基于机器学习和眼动追踪的这类自闭症早期辅助诊断系统真正应用到临床的可行性。
Early diagnosis of autism spectrum disorder (ASD) is very important for improving autism treatment.Recent studies have investigated early diagnosis of children with ASD using machine learning and eye tracking.This paper presents an eye tracking and pupillary response feature extraction method with a naive Bayes classification model for autism that was tested on the Autism Detection Dataset,a dataset of 25 children with ASD and 50 children with typical development aged 3-6 to identify abnormal pupillary responses in children with autism.The method has an average classification accuracy of 90.67% and an average AUC of 92.24% while using only the pupillary features for modeling,which is better than the 82.2% average accuracy achieved by a pupillary and gaze behavior feature model and 78% average accuracy achieved by a gaze behavior and kinematic feature model.This method is simple and accurate.The results show the effectiveness of this method and the feasibility of real clinical applications of this type for early autism diagnosis based on machine learning and eye tracking.
[1] ROBEL L. Clinical features in autism[J]. Archives de Pédiatrie, 2009, 16(11):1507-1512.
[2] BILDT A DE, SYTEMA S, KETELAARS C, et al. Interrelationship between autism diagnostic observation schedule-generic (ADOS-G), autism diagnostic interview-revised (ADI-R), and the diagnostic and statistical manual of mental disorders (DSM-IV-TR) classification in children and adolescents with mental retardation[J]. Journal of Autism and Developmental Disorders, 2004, 34(2):129-137.
[3] 张琴. DSM-Ⅳ孤独症诊断标准与克氏行为量表对儿童孤独症的诊断评价[J].医学理论与实践, 2006, 19(5):586-587.ZHANG Q. The diagnosis and evaluation of DSM-Ⅳ autism criteria and Klinefelter Behavior Scale for children with autism[J]. The Journal of Medical Theory and Practice, 2006, 19(5):586-587.(in Chinese)
[4] 邬方彦,徐秀,刘静,等.孤独症筛查量表(CHAT-23)的应用研究[J].中国儿童保健杂志, 2010, 18(4):288-291. WU F Y, XU X, LIU J, et al. Study on the application of Autism Screening Scale (CHAT-23)[J]. Chinese Journal of Child Health Care, 2010, 18(4):288-291.(in Chinese)
[5] THABTAH F, PEEBLES D. A new machine learning model based on induction of rules for autism detection[J]. Health Informatics Journal, 2020, 26(1):264-286.
[6] MAO Y, HE Y, LIU L, et al. Disease classification based on eye movement features with decision tree and random forest[J]. Frontiers in Neuroscience, 2020, 14:798.
[7] VABALAS A, GOWEN E, POLIAKOFF E, et al. Applying machine learning to kinematic and eye movement features of a movement imitation task to predict autism diagnosis[J]. Scientific Reports, 2020, 10:8346.
[8] BAST N, BANASCHEWSKI T, DZIOBEK I, et al. Pupil dilation progression modulates aberrant social cognition in autism spectrum disorder[J]. Autism Research, 2019, 12(11):1680-1692.
[9] NYSTRÖM P, GLIGA T, JOBS E N, et al. Enhanced pupillary light reflex in infancy is associated with autism diagnosis in toddlerhood[J]. Nature Communications, 2018, 9(1):1678-1682.
[10] HALL C A, CHILCOTT R P. Eyeing up the future of the pupillary light reflex in neurodiagnostics[J]. Diagnostics, 2018, 8(1):19-38.
[11] DICRISCIO A S, HU Y, TROIANI V. Brief report:Pupillometry, visual perception, and ASD features in a task-switching paradigm[J]. Journal of Autism and Developmental Disorders, 2019, 49(12):5086-5099.
[12] DICRISCIO A S, HU Y, TROIANI V. Brief report:Visual perception, task-induced pupil response trajectories and ASD features in children[J]. Journal of Autism and Developmental Disorders, 2019, 49(7):3016-3030.
[13] DINALANKARA D M R, MILES J H, TAKAHASHI N, et al. Atypical pupillary light reflex in 2-6-year-old children with autism spectrum disorders:Pupillary light reflex in 2-6 years old[J]. Autism Research, 2017, 10(5):829-838.
[14] FAN X, MILES J H, TAKAHASHI N, et al. Sex-specific lateralization of contraction anisocoria in transient pupillary light reflex[J]. Investigative Opthalmology&Visual Science, 2009, 50(3):1137.
[15] DICRISCIO A S, TROIANI V. Pupil adaptation corresponds to quantitative measures of autism traits in children[J]. Scientific Reports, 2017, 7(1):6476-6484.
[16] DICRISCIO A S, HU Y, TROIANI V. Task-induced pupil response and visual perception in adults[J]. PLOS ONE, 2018, 13(12):e0209556.
[17] AGUILLON-HERNANDEZ N, MOFID Y, LATINUS M, et al. The pupil:A window on social automatic processing in autism spectrum disorder children[J]. Journal of Child Psychology and Psychiatry, 2020, 61(7):768-778.
[18] DICRISCIO A S, TROIANI V. Resting and functional pupil response metrics indicate features of reward sensitivity and ASD in children[J]. Journal of Autism and Developmental Disorders, 2021, 51(7):2416-2435.
[19] SARGSYAN D, JAGANNATHA S, MANYAKOV N V, et al. Feature selection with weighted importance index in an autism spectrum disorder study[J]. Statistics in Biopharmaceutical Research, 2019, 11(2):118-125.
[20] 何旭.面向自闭症早期诊断的眼动分析系统[D].北京:清华大学, 2019. HE X. Eye movement analysis system for early diagnosis of autism[D]. Beijing:Tsinghua University, 2019.(in Chinese)
[21] AMERICAN P. Diagnostic and statistical manual of mental disorders[J]. Encyclopedia of the Neurological Sciences, 1994, 25(2):4-8.
[22] SAVITZKY A, GOLAY M J E. Smoothing and differentiation of data by simplified least squares procedures.[J]. Analytical Chemistry, 1964, 36(8):1627-1639.
[23] FAN X, MILES J H, TAKAHASHI N, et al. Abnormal transient pupillary light reflex in individuals with autism spectrum disorders[J]. Journal of Autism and Developmental Disorders, 2009, 39(11):1499-1508.
[24] KOHAVI R, JOHN G H. Wrappers for feature subset selection[J]. Artificial Intelligence, 1997, 97(1-2):273-324.
[25] BATTITI R. Using mutual information for selecting features in supervised neural net learning[J]. IEEE Transactions on Neural Networks, 1994, 5(4):537-550.
