Table 5.

Summary of recent and representative studies aiming to distinguish individuals with ASD from TD individuals using multivariate analysis of functional MRI data. 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
Deshpande et al. (2013)199	15 adolescents and young adults with ASD and 15 TD controls	Gathered fMRI data to study causal connectivity among different brain regions relating to Theory of Mind	19 features related to effective connectivity paths	SVM	Classified participants with maximum 96% accuracy, 97% sensitivity, and 95% specificity
Uddin et al. (2013)200	20 children with ASD and 20 TD children	Collected rs-fMRI and structural MRI data, then identified ten connectivity components associated with functional brain networks	Salience network connectivity features	Logistic regression	Achieved 75% sensitivity and 80% specificity with leave-one-out cross-validation; also validated on an independent cohort
Plitt et al. (2015)201	59 young adults with ASD and 59 TD controls; replication set with 89 ASD and 89 TD controls	Collected rs-fMRI data and defined three sets of regions of interest to create three unique correlation matrices for participants’ time series	Destrieux atlas set describing 162 regions	Radial basis function kernel SVM, among others	Observed a maximum 77% accuracy with leave-one-out cross-validation (among other methods); results did not improve in replication set
Chanel et al. (2016)113	15 adults with ASD and 14 TD adults	Gathered fMRI data to study attention/emotions of participants during static faces and dynamic bodies tasks	Features from dynamic body experiment	SVM	Classified with maximum 92% sensitivity and 92% specificity with leave-one-out cross-validation
Yahata et al. (2016)202	74 adults with ASD and 107 TD adults; 44/27 individuals with ASD and 44/27 TD controls in validation sets 1/2	Evaluated functional connectivity from rs-fMRI; also examined generalizability to other disorders	16 out of 9730 functional connections	Logistic regression	Achieved 85% accuracy with leave-one-out cross-validation; validated with 75% and 70% accuracies in independent cohorts
Emerson et al. (2017)203	11 (48) infants at high risk for ASD with (without) a later diagnosis of ASD	Computed features of functional connectivity from rs-fMRI at 6 months to predict ASD diagnosis at 24 months	59 sets of features (one for each fold of leave-one-out cross-validation)	SVM	Predicted future diagnosis with 82% sensitivity and 100% specificity using leave-one-out cross-validation