Artificial intelligence and machine learning in cardiovascular computed tomography

Table 1 Type of machine learning

Types of machine learning	Function	Examples
Supervised learning (55)	Contains labels and outcomes, deduces inferences for prediction purpose	Includes logistic regression, ridge regression, elastic net regression, Bayesian and artificial neural networks
Unsupervised learning (55)	No labels, independently detects significant relationships.	Includes hierarchical clustering, k- means clustering, principal component analysis
Semi-supervised learning (55)	Properties of both supervised and unsupervised learning	Utilized in image and speech recognition
Re-enforcement learning (55)	Utilizes reward function to execute tasks	Utilized in medical imaging, analytics, and prescription selection

Table 2 Machine learning studies in computed tomography

Ref.	ML approach	Brief study description
ML derived CAC assessment
Al’Aref et al[24]	Multiple ML algorithm	To use CAC and clinical factors for CAD prediction
Tesche et al[26]	ML algorithm	To compare ML derived CT FFR and CAC in CT
Kay et al[27]	ML algorithm	To identify phenotypes of left ventricular hypertrophy in combination with CAC
ML derived CT FFR assessment
Zhou et al[31]	Multiple ML algorithms	To employ CT FFR for myocardial bridge formation prediction
Tang et al[32]	ML algorithm	To compare ML CT FFR, CTA and invasive angiography
Coenen et al[33]	Supervised learning	To identify CAD
ML derived evaluation of plaque characteristics
Dey et al[34]	ML algorithm	To generate ML derived scores from plaque characteristics
Hell et al[35]	ML algorithm	To predict cardiac death from plaque characteristics from CTA
ML derived evaluation of epicardial adipose tissue
Rodrigues et al[38]	ML algorithm	To segment and distinguish between different varieties of EAT
Commandeur et al[39]	Deep learning	To quantify EAT in CT
Otaki et al[40]	Supervised learning	To assess the relationship between EAT in CT and MFR in PET
Miscellaneous applications of ML in CT
Baskaran et al[41]	Deep learning	To assess automatic and manual assessment of left and right cardiac structure and function
Al’Aref et al[42]	Supervised learning	To identify culprit coronary lesions in CT
Beecy et al[43]	Deep learning	To detect acute ischemic stroke in CT
Oikonomou et al[44]	Supervised learning	To utilize perivascular fat for cardiac risk prediction
Eisenberg et al[45]	Deep learning	To evaluate epicardial tissue for MACE events

CT: Computed tomography; CTA: Computed tomography angiography; CAC: Coronary artery calcium; ML: Machine learning; CAD: Coronary artery disease; FFR: Fractional flow reserve; EAT: Epicardial adipose tissue.

Table 3 Big data utilization by machine learning in computed tomography

Ref.	ML approach	Number	Brief study description
Motwani et al[47]	Supervised Learning	10030	To predict 5-yr mortality from CT
Rosandael et al[48]	Supervised Learning	8844	To predict major cardiac events from CT
Han et al[49]	ML algorithm	86155	To predict all-cause mortality from CT

CT: Computed tomography; ML: Machine learning.