Table 1.
Studies of AI’s Application in Left Ventricular Systolic Function—LVEF.
| Authors | Year | Task | Model | Dataset | Results |
|---|---|---|---|---|---|
| Leclerc S. et al. [16] | 2019 | LV segmentation | U-Net | 500 subjects | Accuracy in LV volumes (MAE = 9.5 mL, r = 0.95). |
| Smistad E. et al. [17] | 2019 | LV segmentation | U-Net | 606 subjects | Accuracy for LV segmentation (DSC 0.776–0.786). |
| Leclerc S. et al. [18] | 2020 | LV segmentation | LU-Net | 500 subjects | Accuracy in LV volumes (MAE = 7.6 mL, r = 0.96). |
| Wei H. et al. [28] | 2020 | LV segmentation | CLAS | 500 subjects | Accuracy for LVEF assessment (r = 0.926, bias = 0.1%). |
| Reynaud H. et al. [29] | 2021 | LVEF assessment | Transformer | 10,030 subjects | Accuracy for LVEF assessment (MAE = 5.95%, R2 = 0.52). |
| Ouyang et al. [19] | 2020 | LVEF assessment | EchoNet-Dynamic | 10,030 subjects | Accuracy for LV segmentation (DSC = 0.92), LVEF assessment (MAE = 4.1%), and HFpEF classification (AUC 0.97). |
| Asch F.M. et al. [20] | 2019 | LVEF assessment | CNN | >50,000 studies | AutoEF values show agreement with GT: r = 0.95, bias = 1.0%, with sensitivity 0.90 and specificity 0.92 for detection of EF less than 35%. |
| Zhang J. et al. [21] | 2018 | LVEF assessment GLS assessment Disease detection | CNN | 14,035 studies | Agreement with GT: for LVEF, MAE = 9.7%; for GLS, and MAE = 7.5% and 9.0% (within 2 cohorts). Disease detection: HCM, Amyloid, and PAH (AUC 0.93, 0.87, and 0.85). |
| Tromp J. et al. [30] | 2022 | LVEF assessment | CNN | 43,587 studies | Accuracy for LVEF assessment (MAE 6–10%). |
| Narang A. et al. [15] | 2021 | LVEF assessment | Caption Guidance | 240 subjects | LV size, function, and pericardial effusion in 237 cases (98.8%) and RV size in 222 cases (92.5%) are of diagnostic quality. |
| Asch F.M. et al. [32] | 2021 | LVEF assessment | Caption Health | 166 subjects (Protocol 1) 67 subjects (Protocol 2) |
Protocol 1: agreement with GT: ICC 0.86–0.95, bias < 2%. Protocol 2: agreement with GT: ICC = 0.84, bias 2.5 ± 6.4%. |
| Tokodi M. et al. [24] | 2020 | Disease detection (HFpEF) | TDA | 1334 subjects | Region 4 relative to 1: HR = 2.75, 95%CI 1.27–45.95, p = 0.01. Correlation of NYHA and ACC/AHA stages with regions: r = 0.56 and 0.67. |
LVEF, left ventricular ejection fraction; LV, left ventricle; RV, right ventricle; MAE, mean absolute error; DSC, dice similarity coefficient; CLAS, Co-Learning of Segmentation and Tracking on Appearance and Shape Level; AUC, area under the receiver operating characteristic curve; CNN, convolutional neural network; GT, ground truth; GLS, global longitudinal strain; HCM, hypertrophic cardiomyopathy; PAH, pulmonary arterial hypertension; HFpEF, heart failure with preserved ejection fraction; HF, heart failure; ICC, intraclass correlation coefficient; TDA, topological data analysis; HR, hazard ratio; CI, confidence interval; NYHA, New York Heart Association; ACC, American College of Cardiology; AHA, American Heart Association.