Table 3.

Summary of recent and representative studies aiming to distinguish individuals with ASD from TD individuals using multivariate analysis of motor skill development and eye gaze/tracking patterns. Reported sample sizes are the numbers used for classification and do not necessarily reflect the study’s total sample size.

Reference	Study Participants	Experimental Methods	Key Features	Multivariate Technique	Key Results
Motor Patterns
Crippa et al. (2015)185	15 children with ASD and 15 TD children	Recorded kinematics data while children performed a reach-to-drop task	Seven kinematic features	SVM	Mean sensitivity/specificity of 82%/89% with leave-one-out cross-validation
Dehkordi et al. (2015)96	35 children with ASD and 16 TD children	Evaluated children’s social and behavioral interactions with a robotic parrot	Six behavioral features	Random forest	Classified with a maximum of 90% accuracy using seven-fold cross-validation
Anzulewicz et al. (2016)98	35 children with ASD and 45 TD children	Recorded kinematic and gesture data from children playing with tablet computers	262 motor features derived from the tablet sensor data	Regularized greedy forest, among other techniques	Achieved a maximum average AUROC of 0.93 with ten repetitions of ten-fold cross-validation
Li et al. (2017)186	14 adults with ASD and 16 TD controls	Derived kinematic parameters from a hand movement imitation task	Nine kinematic parameters (from two imitation conditions)	SVM, among others	Achieved 87% accuracy, 86% sensitivity, and 88% specificity using a two-step cross-validation method
Moradi et al. (2017)97	25 children with ASD and 25 TD children	Evaluated movement characteristics of children playing with a smart toy car	Five movement characteristics	Polynomial kernel SVM	Averaged 93% sensitivity and 76% specificity with five-fold cross-validation
Eye Gaze/Tracking
Stahl et al. (2012)187	19 high-risk infants with a sibling with ASD, 17 control infants with no ASD in family	Recorded EEG and measured event-related potentials associated with eye gaze processing	36 event-related potential (18 direct gaze, 18 averted gaze) metrics	SVM	Classified high-risk versus control with 64% sensitivity and 64% specificity
Fujioka et al. (2016)188	21 adolescents and adults with ASD and 35 TD controls	Measured percentage of eye fixation time on objects displayed on a screen	Discrimination parameters from three visual areas of interest	Discriminant analysis	Classified with 81% sensitivity and 80% specificity
Liu et al. (2016)189	29 children with ASD and 58 TD children	Analyzed children’s eye movements during a facial recognition task	Histograms of visual attention to partitioned facial regions	Radial basis function kernel SVM	With leave-one-out cross-validation, achieved 89% accuracy, 93% sensitivity, and 86% specificity
Frazier et al. (2018)190	91 youth diagnosed with ASD and 110 non-ASD youth	Recorded eye tracking patterns of participants while viewing a video containing 44 visual stimuli	Gaze metrics correlating significantly with ASD diagnosis	Multiple linear regression with ROC analysis	Achieved AUROC of 0.92 and 0.86 in the training set (75% of samples) and validation set (25%)
Wan et al. (2018)191	37 children with ASD and 37 TD children	Measured children’s fixation time on ten areas of interest while watching a short video of a young female speaking	Fixation time on the body and mouth	SVM	Classified with 85% accuracy, 87% sensitivity, and 84% specificity