Table 4. Analysis of translation efforts.

AUC: Area under the curve; CTA: Computed tomography angiography; ICP: Intracranial pressure; MAP: Mean arterial pressure; CSF: Cerebrospinal fluid; LVO: Large vessel occlusion; IB: Imaging biometrics; GBM: Glioblastoma; CT: Computerized tomography; MRI: Magnetic resonance imaging; BMI: Body mass index

Study/device	Externally validated	Applied for FDA approval	Translated into clinic	Type of model	Training method	Input	Outcome	Model performance
Valliani et al.[38]	Yes	No	No	Ensemble learning algorithm	Single center	Clinical features and surgical variables	Non-home discharge after thoracolumbar spine surgery	AUC: 0.81 (test) AUC: 0.77 (external)
Senders et al.[39]	Yes	No	No	Accelerated failure time model	Retrospective dataset	Demographic, socioeconomic, clinical, and radiographic features	Prediction of survival in glioblastoma patients	C-index: 0.70 (best time-to-event model), 0.70 (best continuous and binary model)
Liu et al.[40]	Yes	No	No	Decision tree	Retrospective	Demographic, clinical, and aneurysm-specific features as well as Glasgow Coma Score	Long-term outcomes after poor-grade aneurysmal subarachnoid hemorrhage	AUC: 0.88 (test) AUC: 0.94 (external)
Staartjes et al.[41]	Yes	No	No	Logistic generalized additive model	Multicenter retrospective database	Clinical features and labs	Functional impairment after intracranial tumor surgery	AUC: 0.72 (test), AUC: 0.72 (external)
Viz Aneurysm[42]	Yes	Yes	Yes	Neural network	Unknown	CTA imaging	Detection and analysis of cerebral aneurysms	93.8% sensitivity 94.2% specificity (external)
Carra et al.[43]	Yes	No	No	Random forest	Prospective and retrospective databases	ICP and MAP signals	Doses of harmful intracranial pressure in patients with severe traumatic brain injury	AUC: 0.94 (external)
Aidoc BriefCase-CSF triage[44]	Yes	Yes	Yes	Neural network	Unknown	CTA images	Detection of cervical spine fractures	(external) Sensitivity - 54.9 Specificity - 94.1
Rapid LVO 1.0[45]	Yes	Yes	Yes	Density threshold-based traditional machine learning	Unknown	CTA images	Detection of anterior circulation LVOs	Sensitivity - 96.0 Specificity - 98.0
Imaging biometrics (IB) Neuro-Oncology[46]	Yes	Yes	Yes	Artificial Intelligence	Unknown	CTA images	Distinguishing glioblastoma progression	Classification of survival in GBM sensitivity - 80.0 Specificity - 63.0
Staartjes et al.[47]	Yes	No	No	Logistic regression	Retrospective multicenter database	Clinical features and surgical variables	Prediction of mid-term outcomes after lumbar spinal fusion	Functional Impairment AUC:0.75 (test) AUC:0.67 (external) Back pain AUC: 0.71 (test) AUC:0.72 (external) Leg Pain AUC: 0.72 (test) AUC: 0.64 (external)
Thanellas et al.[48]	Yes	No	No	Convolutional neural network	Retrospective single-center dataset	CTA images	Identification and localization of subarachnoid hemorrhage on CT scans	Development: Sensitivity 87.4% and specificity 95.3% External: sensitivity 99.3% and specificity 63.2%
Teng et al.[49]	Yes	No	No	Logistic regression	Retrospective multicenter database	Clinical features lab values and imaging	Predicting high-grade intracranial meningioma	Development: AUC: 0.99 AUC: 0.75 (internal) AUC: 0.842 (external)
Vitrea CT Brain Perfusion[50]	Yes	Yes	Yes	Convolutional neural network	Unknown	CTA images	Detection of intracranial aneurysms	Sensitivity - 90.9-96.3 Specificity - 100.0
Ma et al.[51]	Yes	No	No	Neural network	Retrospective multicenter database	MRI images	Prediction of intramedullary glioma grade	AUC: 0.8431
Biswas et al.[52]	Yes	No	No	Neural network	Retrospective single-center database	Predictor variables (demographic and lab)	Predicting chronic subdural hematoma referral outcome	AUC: 0.951 (test) AUC: 0.896 (external)
Fang et al.[53]	Yes	No	No	Convolutional neural network	Retrospective single-center database	Demographic, clinical, and imaging variables	Glasgow Outcome Scale	AUC (test): 0.93 AUC (external): 93.69
Karhade et al.[54]	Yes	No	No	Logistic regression	Retrospective multicenter database	Clinical labs and tumor-specific variables	Six-week mortality in spinal metastasis cases	AUC (test): 0.84 AUC (external): 0.81
Crabb et al.[55]	Yes	No	No	Logistic regression, random forest, support vector machine, and ensemble learning algorithm	Public database	Age, BMI, clinical variables, and preoperative lab tests	Prediction of unplanned 30-day readmissions after pituitary adenoma resection	AUC (external): 0.76
Warman et al.[56]	Yes	No	No	Ensemble learning algorithm	Prospective single center dataset and public dataset	Patient demographics, presenting vital signs, mechanism of injury, initial Glasgow Coma Scale (GCS)	Predicting in-hospital mortality after traumatic brain injury	AUC (test): 0.91 AUC (external): 0.89
Habets et al.[57]	Yes	No	No	Logistic regression	Retrospective single center	Patient demographic data, disease-specific data, clinical performance scores, and neuropsychological scores	Prediction of motor response after deep brain stimulation	AUC (test): 0.79 AUC (external) 0.79