Estimating Ejection Fraction from Single View Echocardiographic Images

Heart failure is a prevalent and clinically significant condition, with ejection fraction (EF) serving as a key parameter for its classification. Given its critical role in diagnosis and patient evaluation, management, precise EF estimation is a priority in cardiovascular imaging for guiding therapeutic decisions. This study introduces an automated framework for EF quantification using a pair of four-chamber view (4CH) echocardiographic images-captured at end-diastolic and end-systolic phases, integrating deep learning and geometric modeling. An attention-based U-Net segments the left ventricle (LV), while an ellipsoidal model enables three-dimensional LV reconstruction for EF calculation. Performance evaluation against ground truth data demonstrates minimal estimation errors, with Bland-Altman analysis confirming strong agreement. The classification of EF groups, conducted to assess the clinical impact of deviations in measurements, yielded an accuracy of 82.0%. Despite these promising results, further refinement of three-dimensional ventricular volume reconstruction is necessary to enhance precision.Clinical relevance- An affordable and accurate EF calculation can support timely heart failure classification and management.