Auto-LVEF: A Novel Method to Determine Ejection Fraction from 2D Echocardiograms

Ejection Fraction (EF) is a primary indicator of one’s heart’s performance but calculating it has been a challenging task which is a very tedious and time-consuming process. Healthcare is a domain that can reap the benefits of AI, but when it comes to this domain, most Artificial Intelligence (AI) models suffer explainability - The Blackbox, though there are several Explainable Artificial Intelligence (XAI) techniques, healthcare is a domain where even an iota of mistake may result in a grave life-threatening predicament. Until we are 100% sure of why the model is working the way it is working, we cannot deploy it for clinical use. Taking all these issues into consideration we have developed a simple yet novel and interpretable method to determine left ventricular ejection fraction from a 2-dimensional Transthoracic Echocardiogram (2D TTE) without any human intervention at any point of the process. Our model was evaluated on the EchoNet-Dynamic dataset and it outperforms current state-of-the-art models with an accuracy of 0.97 measured in \(R^2\) while still being a white box model.