Table 2.
Examples of machine learning methods applied in neurodevelopmental paediatrics.
| Authors | Das and Khanna75 | Sen et al.76 | Emerson et al.79 | Girault et al.80 | Kim et al.81 |
| Year published | 2019 | 2018 | 2017 | 2019 | 2021 |
| Purpose | Diagnostic | Diagnostic | Prognostic | Prognostic | Diagnostic |
| Outcome | ADHD | ADHD | ASD at 24 months of age | Cognitive ability at age 2 years | Developmental disability |
| Outcome measure | Diagnosed by child neurologist using DSM-IV criteria | Diagnostic criteria differed across study sites | DSM-IV criteria | Mullens Scales of Early Ability. Outcome categorised as above or below the median | Diagnosis made by paediatric psychologist |
| Features | Pupillometric data | Structural and functional MRI | MRI at 6 months of age | MRI at birth | Drag and drop data from computer-based game |
| Participants | 50 (28 cases, 22 controls) | 558 (279 cases, 279 controls) | 59 infants with high familial risk of ASD (11 cases, 48 controls) | 115 infants | 370 children (147 cases, 233 controls) |
| Algorithms trained | LR, SVM, KNN, NB, DT, RF | SVM | SVM | DL | DL |
| Evaluation method | Nested 10-fold cross-validation | Independent test set (77 cases, 94 controls) | Cross-validation with nested leave one out procedure | Independent test set (22 cases, 15 controls) | 10-fold cross-validation |
| Evaluation result (best performing models) | NB: AUROC 0.871, sensitivity 0.656, specificity 0.863 | Accuracy 0.6725, sensitivity 0.4545, specificity 0.8510 | Accuracy 0.966, sensitivity 0.818, specificity 1.000 | Accuracy 0.838, sensitivity 0.864, specificity 0.800 | AUROC 0.817, sensitivity 0.757, specificity 0.740 |
LR logistic regression, SVM support vector machine, KNN K-nearest neighbours, NB naive Bayes, DT decision tree, RF random forest, ADHD attention deficit hyperactivity disorder, ASD autism spectrum disorder, DSM-IV diagnostic and statistics manual of mental disorders.