Table 2.
Summary of studies mentioned in review.
| Publication | Indication | Type of machine learning algorithm | Size of dataset | Performance of model tested | Tested on prospective dataset |
|---|---|---|---|---|---|
| Claassen et al. (25) | Detection of cognitive-motor dissociation on EEG | SVM | 240 EEGs from 104 patients | Significantly improved prognosis of neurological recovery (OR 4.6; CI 1.2–17.1) | Yes |
| Raj et al. (27) | Predict 30-day all-cause mortality after TBI with invasive ICP measurements | Modified logistic regression using dynamic variables | 472 patients | AUC = 0.84 | No |
| Arbabshirani et al. (28) | Detection of intracranial hemorrhage | CNN | 46,583 CT scans | AUC = 0.85 | Yes |
| Tanioka et al. (31) | Prediction of DCI after subarachnoid hemorrhage | Random Forest | 95 patients | Prediction accuracy = 95.1% | Yes |
| Struck et al. (34) | Seizure prediction using cEEGs | RiskSLIM | 7,716 cEEGs | AUC = 0.83 | Yes |
| Koren et al. (35) | Assist EEG reviewers to annotate different cEEG patterns | Neurotrend | 76 cEEGs | Multi-rater agreement for burst suppression (Gwet's coefficient = 0.86) | No |
| Yu et al. (37) | Predict hemorrhagic transformation | Kernel spectral regression | 165 patients | AUC = 0.84 | No |
| Savin et al. (39) | Prediction of healthcare-associated ventriculitis and meningitis | XGBoost | 2,286 patients | AUC = 0.83 | No |
| Stapleton et al. (42) | Identification of metabolite associated with neurological outcomes following subarachnoid hemorrhage | LASSO Regression | 137 patients | Found plasma levels of taurine were 21.9% higher in patients with good vs. poor outcomes (P = 0.002) | No |
| Hernandes Rocha et al. (44) | Predict neurological recovery following TBI | Bayesian generalized linear model | 3,138 patients | AUC = 0.87 | No |
| Tabrizi et al. (46) | Predict post-hemorrhagic hydrocephalus outcomes in neonates with intraventricular hemorrhage using cranial ultrasound | SVM | 64 patients | Prediction Accuracy = 84% | No |
| Heaphy-Henault et al. (47) | Predict congenital aqueductal stenosis with fetal MRI and the most important fetal MRI findings associated with congenital aqueductal stenosis | Random Forest | 75 patients | Found enlarged inferior recesses of the third ventricle were the most important fetal MRI features associated with congenital aqueductal stenosis (P < 0.0023) | No |
| Pisapia et al. (48) | Predicting which patients would require post-natal cerebrospinal fluid diversion with fetal MRI | SVM | 253 | Predictive accuracy = 82% on initial cohort, and 91% on independent cohort | No |
Table shows a summary of the different studies mentioned in this review.