Table 1 Previous AI studies of echocardiography
From: Artificial intelligence in echocardiography: detection, functional evaluation, and disease diagnosis
| Â | Aim of study | AI algorithm or commercial technique | Detail method/software | Diagnostic Efficiencya | Reference |
|---|---|---|---|---|---|
Section recognition/ image acquisition | Automatic classification of cardiac views | AI algorithm | CNN | ACC = 98.3% | Ultrasound Med Biol, 2019 [19] |
AI algorithm | CNN | ACC = 97.8% | NPJ Digital Med, 2018 [12] | ||
Segmentation | Left ventricular segmentation (provide accuracy improvement of the endocardial boundary recognition) | AI algorithm | Ant colony optimization | N/A | Biomed Mater Eng, 2014 [13] |
AI algorithm | Radial active contour method Snake | N/A | Comput Methods Programs Biomed, 2014 [20] | ||
AI algorithm | Iterative 3-D cross-correlation algorithm | N/A | Ultrasound Med Biol, 2014 [21] | ||
AI algorithm | Ant colony optimization | N/A | Biomed Mater Eng, 2014 [13] | ||
Right ventricular segmentation | AI algorithm | Sparse matrix transform/ wall thickness constraint feature | Dice = 90.8% (epicardial boundary), 87.3% (endocardial boundary) | Phys Med Biol, 2013 [22] | |
Atrial and multi-chamber heart segmentation | AI algorithm | Active shape model/ fusion-imaging technology | Mean dice = 83.3% ~ 91.3% (LV) | Ultrasound Med Biol, 2015; IEEE Trans Ultrason Ferroelectr Freq Control, 2015 [23, 24] | |
LV assessment | Assessment of left heart function | Commercial technique | HeartModel (Philips) | r = 0.87 ~ 0.96/ r = 0.98 | JACC: Cardiovascular Imaging, 2016; J Am Soc Echocardiogr, 2017 [25, 26] |
Commercial technique | AutoLV (TomTec Imaging system) | ICC = 1.0 | J Am Coll Cardiol, 2015 [27] | ||
Commercial technique | AutoEF (BayLabs) | r = 0.95 | Circ Cardiovasc Imaging, 2019 [28] | ||
AI algorithm | 3D CNN | AUC = 0.92 | J Am Soc Echocardiogr, 2020 [29] | ||
AI algorithm | PWC-Net | ACC = 97% ~ 98% | JACC Cardiovasc Imaging, 2021 [30] | ||
AI algorithm | Neural network | Intraclass correlation = 0.86 – 0.95 (AI and physician), 0.84 (novice using AI and physician) | Circ Cardiovasc Imaging, 2021 [31] | ||
Cardiac disease diagnosis | Valvular heart disease | Commercial technique | Proximal isovelocity surface area (PISA) | Intraclass correlation coefficients = 0.96 | Circ Cardiovasc Imaging, 2013; J Am Soc Echocardiogr, 2012 [32, 33] |
Commercial technique | Real-time 3D volume color-flow Doppler (RT-VCFD) | r = 0.93 | Int J Cardiovasc Imaging, 2015 [34] | ||
Commercial technique | Mitral Valve Navigator (Philips) | N/A | Echocardiography, 2016 [35] | ||
Commercial technique | Anatomical Intelligence in ultrasound (AIUS) | ACC = 89% | J Am Soc Echocardiogr, 2016 [36] | ||
AI algorithm | Anatomical affine optical flow | Intra- and interobserver variability = 0.85 and 0.65 | Int J Cardiovasc Imaging, 2019 [37] | ||
AI algorithm | 2D/3D CNN | Sensitivity (Recall) = 67%, PPV = 74% ~ 77% | Nat Commun, 2021 [38] | ||
Cardiomyopathy | Commercial technique | Myocardial strain analysis | N/A | Circ J, 2019 [39] | |
AI algorithm | Associative memory classifier | AUC = 89.2% | Circ Cardiovasc Imaging, 2016 [40] | ||
AI algorithm | Support Vector Machine | AUC = 77.8% | AMIA Annu Symp Proc, 2014 [41] | ||
Coronary atherosclerotic heart disease | AI algorithm | discrete wavelet transform and texture feature analysis | AUC = 90% ~ 99% | JACC Cardiovasc Imaging, 2019 [42] | |
Commercial technique | EchoPAC PC (GE) | r = 0.67 ~ 0.99 | Echocardiography, 2015 [43] | ||
AI algorithm | A specialised software for MCE quantification | Sensitivity = 77% Specificity = 94% | Ultrasound Med Biol, 2017 [15] | ||
AI algorithm | Texture analysis | Highest accuracy = 79% | IEEE, 2016 [44] | ||
Intracardiac masses | AI algorithm | Feed-forward neural network | Accuracy = 91% | Comput Med Imaging Graph, 2006 [45] | |
AI algorithm | Gray level co-occurrence matrix-based features/ ANN (Classification) | Mean sensitivity = 0.955 Specificity = 0.970 | J Ultrasound Med, 2014 [46] |