Table 2.

Deep-learning-based AI studies for view identification and quality assessment. MAE: mean absolute error.

	Task	DL Model	Data/Validation	Performance
Zhang et al. [16,17]	23 standard echo view classification	Customized 13-layer CNN model	5-fold cross validation/7168 cine clips of 277 studies	Overall accuracy: 84% at individual image level
Mandani et al. [19]	15 standard echo view classification	VGG [26]	Training: 180,294 images of 213 studies Testing: 21,747 images of 27 studies	Overall accuracy: 97.8% at individual image level and 91.7% at cine-lip level
Akkus et al. [20]	24 Doppler image classes	Inception_resnet [27]	Training: 5544 images of 140 studies Testing: 1737 images of 40 studies	Overall accuracy of 97%
Abdi et al. [21,22]	Rating quality of apical 4 chamber views (0–5 scores)	A customized fully connected CNN	3-fold cross validation/6196 images	MAE: 0.71 ± 0.58
Abdi et al. [23]	Quality assessment for five standard view planes	CNN regression architecture	Total dataset: 2435 cine clips Training: 80% Testing: 20%	Average of 85% accuracy
Dong et al. [24]	QC for fetal ultrasound cardiac four chamber planes	Ensembled three CNN model	5-fold cross validation (7032 images)	Mean average precision of 93.52%.
Labs et al. [25]	Assessing quality of apical 4 chamber view	Hybrid model including CNN and LSTM layers	Training/validation/testing (60/20/20%) of in total of 1039 images	Average accuracy of 86% on the test set