**Authors and Affiliations**
Anisuzzaman, D.M., Malins, J.G., Jackson, J.I., Lee, E., Naser, J.A., Rostami, B., Bird, J., Friedman, P.A., Lopez-Jimenez, F., Pislaru, S.V., Kane, G.C., Attia, Z.I.

Affiliations: Department of Cardiovascular Medicine, Mayo Clinic, Rochester, MN

**Background**
Recently, our team developed an end-to-end artificial intelligence (AI) framework that can successfully estimate left ventricular ejection fraction (LVEF) from specific echocardiogram views (A2C and PLAX) yet cannot estimate LVEF for studies from which these views are missing. In this project, we developed a deep learning model to estimate LVEF from any valid (any B-mode echocardiogram video containing left ventricle) echocardiogram video(s).

**Methods**
We built a deep learning model using retrospective transthoracic echocardiography (TTE) data, which can take valid echocardiogram video(s) as input. The model was evaluated with two test datasets: one with all available valid clips, and one with preselected views (both having 433 patients). Later, the model was tested with a prospective patient cohort that had both TTE and point-of-care ultrasound (POCUS) data collected simultaneously (565 patients).

**Results**
On the retrospective TTE test set, we observed excellent performance (all valid clips: RMSE of 6.62% and Pearson correlation (PC) of 0.84 on LVEF estimation, and an AUC of 0.96 (95% CI: of 0.95-0.98) on low LVEF classification (<=40%); preselected views: RMSE of 6.75%, PC of 0.83, and an AUC of 0.96 (95% CI: 0.94-0.98)). For both TTE and POCUS data from the prospective cohort, we observed similar performance amongst the two types of images.

**Conclusion**
To the best of our knowledge, this is the first view-invariant, end-to-end deep learning model developed to estimate LVEF. The developed model can be particularly useful when a POCUS exam is more feasible (e.g., in an emergency department; for pre-screening; at the bedside for critically ill patients). The model also showed superior performance compared to our previous view-specific model. Furthermore, the developed model holds some other advantages such as not requiring volume tracing or manual view selection.



**Authors and Affiliations**
_Anisuzzaman, D.M., Malins, J.G., Jackson, J.I., Lee, E., Naser, J.A., Rostami, B., Bird, J., Friedman, P.A., Lopez-Jimenez, F., Pislaru, S.V., Kane, G.C., Attia, Z.I._

**Affiliations:** Department of Cardiovascular Medicine, Mayo Clinic, Rochester, MN

**Background**
Recently, our team developed an end-to-end artificial intelligence (AI) framework that can successfully estimate left ventricular ejection fraction (LVEF) from specific echocardiogram views (A2C and PLAX) yet cannot estimate LVEF for studies from which these views are missing. In this project, we developed a deep learning model to estimate LVEF from any valid (any B-mode echocardiogram video containing left ventricle) echocardiogram video(s).

**Methods**
We built a deep learning model using retrospective transthoracic echocardiography (TTE) data, which can take valid echocardiogram video(s) as input. The model was evaluated with two test datasets: one with all available valid clips, and one with preselected views (both having 433 patients). Later, the model was tested with a prospective patient cohort that had both TTE and point-of-care ultrasound (POCUS) data collected simultaneously (565 patients).

**Results**
On the retrospective TTE test set, we observed excellent performance (all valid clips: RMSE of 6.62% and Pearson correlation (PC) of 0.84 on LVEF estimation, and an AUC of 0.96 (95% CI: of 0.95-0.98) on low LVEF classification (<=40%); preselected views: RMSE of 6.75%, PC of 0.83, and an AUC of 0.96 (95% CI: 0.94-0.98)). For both TTE and POCUS data from the prospective cohort, we observed similar performance amongst the two types of images.

**Conclusion**
To the best of our knowledge, this is the first view-invariant, end-to-end deep learning model developed to estimate LVEF. The developed model can be particularly useful when a POCUS exam is more feasible (e.g., in an emergency department; for pre-screening; at the bedside for critically ill patients). The model also showed superior performance compared to our previous view-specific model. Furthermore, the developed model holds some other advantages such as not requiring volume tracing or manual view selection.



A View-Invariant Deep Learning Model for Estimating Left Ventricular Ejection Fraction from Any Valid Echocardiogram Videoclip, Including Point-Of-Care Ultrasound (POCUS)