Table 2.

CVD detection based on multi-modal dataset.

Ref.	Year	Country	Fused Data	Results	Summary
[50]	2019	USA	Electronic Health Record (EHR) and Genetic data	AUROC: 0.790 AUPRC 0.285	The study used a 10-year data from HER and genetic data to predict CVD events using random forest, gradient boosting trees, logistic regression, CNN and long short-term memory (LSTM). Chi-squared was used for feature selection on the EMR data. Results show an improved prediction of CVD with AUROC of 0.790 compared to EMR alone (AUC of 0.71) or genetic alone (AUC of 0.698)
[51]	2020	China	Electrocardiogram (ECG), Phonocardiogram (PCG), Holter monitoring, Echocardiography (ECHO), and biomarker levels (BIO)	Accuracy: 96.67 Sensitivity: 96.67 Specificity: 96.67 F1-score: 96.64	The study aimed at the detection of coronary artery disease (CAD). Data from ECG and PCG of 62 patients were used. Furthermore, data were also collected from Holter monitoring, ECHO and BIO. Feature selection was applied to attain optimum features and support vector machine was used for classification. Results show best performance when feature were fused from all sources.
[52]	2020	USA	Sensors (collect blood pressure, oxygen, respiration rate, etc.) and Medical records (history, lab test, etc.)	Results after feature weighting method: Accuracy: 98.5 Recall: 96.4 Precision: 98.2 F1-score: 97.2 RMSE: 0.21 MAE: 0.12	The study aimed at predicting heart diseases (such as heart attack or stroke) using data gathered from sensors and medical records. Features such as age, height, BMI, respiration rate, and blood pressure were extracted, and then data from both sources were fused. Furthermore, conditional probability is utilized for feature weighting to help in accuracy improvement. An ensemble deep learning is then used for the prediction of heart disease.
[53]	2021	Greece	Myocardial Perfusion Imaging (MPI) and Clinical data	Accuracy: 78.44 Sensitivity: 77.36 Specificity: 79.25 F1-score: 75.50 AUC: 79.26	The study aimed at cardiovascular disease diagnosis using MPI and Clinical data. Polar maps were derived from the MPI data and fused with clinical data of 566 patients. Random forest, neural network, and deep learning with Inception V3 were used for classification. Results show a hybrid model of Inception V3 with random forest achieved an accuracy of 78.44% compared to an accuracy of 79.15% achieved by medical experts.
[54]	2021	USA	Electronic medical records (EMR) and Abdominopelvic CT imaging	AUROC: 0.86 AUCPR: 0.70	The study aimed at developing a risk assessment model of ischemic heart disease (IHD) using combined information from patientsí EMR and features extracted from abdominopelvic CT imaging. In this study, CNN used to extract features from images and XGBoost was used as the learning algorithm. Results show an improved prediction performance with AUROC of 0.86 and AUCPR of 0.70
[55]	2021	USA	Genetic, clinical, Demographic, imaging, and lifestyle.	-	The study aimed at evaluating the ability of machine learning in detecting CAD subgroups using multimodal data. The multimodal data consisted of genetic, clinical, demographic, imaging, and lifestyle data. K-means clustering as well as Generalized low rank Modeling were utilized. Results show that 4 subgroups were uniquely identified.