Table 2. Comparison of the studies that performed the ML vs. non-ML CVD/stroke risk assessment.

Author (year)	N	ML algorithm	CVRCStat	Imaging modality	CCVRF + image phenotypes	#F	Types of features	Ground truth	Training protocol	FU (years)	PE
Narain et al. (105), 2016	689	QNN	FRS	×	×	7	CCVRF	From physician	K3	–	ACC: 98.57%
Unnikrishnan et al. (50), 2016	2,406	SVM	FRS	×	×	9	CCVRF	CVD events	K5	10	AUC: 0.71
Weng et al. (26), 2017	378,256	RF, LR, GBM, & ANN	ASCVD	×	×	30	CCVRF	CVD events	K4	10	AUC: 0.764
Venkatesh et al. (49), 2017	6,814	RF	FRS	MRI + CUS	√	735	CCVRF, image phenotypes, & serum biomarkers	CVD events	K3	12	C-index: 0.81
Zarkogianni et al. (51), 2017	560	HWNN	UKPDS	×	×	15	CCVRF	CVD events	K10	5	AUC: 0.71
Kakadiaris et al. (48), 2018	6,459	SVM	ASCVD	×	×	9	CCVRF	CVD events	K2	13	AUC: 0.92
Han et al. (106), 2019	86,155	Boosted ensemble	FRS & ASCVD	CT	√	70	Clinical + laboratory + CAC	CVD events	Hold-out	4.6	AUC: 0.82
Proposed, 2019	202	SVM	13 CVRC	CUS	√	19	CCVRF + CUSIP	EEGS	K10	–	AUC: 0.70 (CCVRF); AUC: 0.88 (CCVRF + CUSIP)

CVD, cardiovascular diseases; QNN, quantum neural network; SVM, support vector machine; RF, random forest; LR, logistic regression; GBM, gradient boosting machines; ANN, artificial neural network; HWNN, hybrid wavelet neural networks; CVRC, cardiovascular risk calculators; CUS, carotid ultrasound; CT, computed tomography; MRI, magnetic resonance imaging; CCVRF, conventional cardiovascular risk factors; ASCVD, atherosclerosis cardiovascular disease; FRS, Framingham risk score; FU, follow-up; PE, performance evaluation; CUSIP, carotid ultrasound image-based phenotypes; AUC, area under the curve.