A mix of Wavelet Transforms and CNN for Left Ventricular Ejection Fraction Estimation

Echocardiography imaging plays a pivotal role in heart failure diagnostic, offering a non-invasive and versatile approach for assessing ejection fraction. However, the accuracy of ejection fraction estimation depends on operator skill and takes time to realize, in an emergency setting this time wasted could be used in more critical tasks. This paper tries to make ejection fraction assessment available to all clinicians even if they aren’t trained for it. For that we use a novel resource efficient approach named WaveMix [1] a neural network based on a multilevel two-dimensional discreet wavelet transform combined with convolutional layers. Our experiments demonstrate that higher DWT levels enhance the model’s accuracy, showing promise for deployment in clinical settings.