. 2018 Oct 29;5(4):R115–R125. doi: 10.1530/ERP-18-0056
This work is licensed under a Table 2.
Basic finding and validation of machine learning applications in the field of echocardiography.
| Study | Year | Application | Machine learning model used | Training/validation set | Test set | Limit of agreements and bias | Sensitivity/Specificity/Accuracy | AUC | Time required for measurement |
|---|---|---|---|---|---|---|---|---|---|
| (7) | 2018 | Recognise 15 echocardiography views | Convolutional neural network | 200,000 images | 20,000 images | – | –/–/91.7% | 0.996 | 21 ms/image |
| (54) | 2018 | Quantification of wall motion abnormalities | Double density-dual tree discrete wavelet transform | 279 images | – | – | 96.12%/96%/96.05% | – | – |
| (55) | 2018 | Quantification of wall motion abnormalities | Convolutional neural network | 4392 maps | 61 subjects | 81.1%/65.4%/75% | – | – | |
| (36) | 2017 | Recognition/classification of apical views | Supervised dictionary learning | 210 clips | 99 clips | – | –/–/95% | – | 0.05 ± 0.003 s per clip |
| (57) | 2017 | Assessment of myocardial velocity | Unsupervised multiple kernel learning | 55 subjects | – | Avg 51.7% | Avg 73.25%/78.4%/– | – | <30 s |
| (5) | 2016 | Classification/discrimination of pathological patterns (HCM vs ATH) | Support vector machine, random forest, artificial neural network | – | – | – | 96%/77%/– | 0.795 | 8 s |
| (27) | 2016 | Classification/discrimination of pathological patterns (RCM vs CP) | Associative memory-based machine-learning algorithm | – | – | – | –/–/93.7% | 0.962 | – |
| (47) | 2016 | Quantification of MR | Support vector machine | 5004 frames | – | – | 99.38%/99.63%/99.45% | – | – |
| (24) | 2015 | Calculation of EF and LS | AutoEF Software | – | 255 patients | 0.83 (0.78 to 0.86) and −0.3 (1.5 to 0.9) | – | – | 8 ± 1 s/patient |
| (37) | 2013 | Automated detection of LV border | Random forest classifier with an active shape model | 50 images | 35 images | – | –/–/90.09% | – | – |
| (53) | 2011 | Quantification of wall motion abnormalities | Relevance Vector Machine classifier | 173 patients | – | – | –/–/93.02% | – | – |
| (56) | 2008 | Quantification of wall motion abnormalities | Hidden Markov model | 24 studies (720 frames) | 20 studies (600 frames) | – | –/–/84.17% | – | – |
| (39) | 2008 | Calculation of EF | AutoEF Software | 10,000 images | 92 patients | 1% (−19% to 33%) | – | – | – |
| (38) | 2007 | Calculation of EF | AutoEF Software | >10,000 images | 200 patients | 6% (−2.87 to 2.91) | – | – | <15 s per view |
ATH, athletes’ heart; Avg, average; CP, constrictive pericarditis; EF, ejection fraction; HCM, hypertrophic cardiomyopathy; LS, longitudinal strain; LV, left ventricle; MR, mitral regurgitation; ms, milliseconds; RCA, restrictive cardiomyopathy; s, seconds.