Table 4.

Deep-learning-based AI studies for disease diagnosis. AUC: area under the curve.

	Task	DL Model	Data/Validation	Performance
Zhang et al. [16,17]	Diagnosis of hypertrophic cardiomyopathy (HCM), cardiac amyloidosis (amyloid), and pulmonary hypertension (PAH)	VGG [26]	HCM: 495/2244 Amyloid:179/804 PAH:584/2487 (Diseased/Control) 5-fold cross validation	Hypertrophic cardiomyopathy: AUC of 0.93; cardiac amyloidosis: AUC of 0.87; pulmonary hypertension: AUC of 0.85
Ghorbani et al. [33]	Diagnose presence of pacemaker leads; enlarged left atrium; LV hypertrophy	A customized CNN model	Training set: 1.6 million images from 2850 patients; test set: 169,000 images from 373 studies	Presence of pacemaker leads with AUC = 0.89; enlarged left atrium with AUC = 0.86, left ventricular hypertrophy with AUC = 0.75.
Ouyang et al. [15]	Predict presence of HF with reduced EF	3D convolutions with residual connection	Training set: 7465 echo videos; internal test dataset (n = 1277); external test dataset (n = 2895)	AUC of 0.97
Omar et al. [35]	Detecting wall motion abnormalities	Modified VGG-16 [26]	120 echo studies. One-leave-out cross validation	Accuracy: RF = 72.1%, SVM = 70.5% CNN = 75.0%
Kusunose et al. [36]	Detecting wall motion abnormalities (WMA)	Resnet [38]	300 patients with WMA +100 normal control. Training = 64% Validation:16% Test: 20%	AUC of 0.99
Narula et al. [37]	Differentiate HCM from ATH	A customized ANN	77 ATH and 62 HCM patients. Ten-fold cross validation	Sensitivity: 87% Specificity: 82%