Volume quantification is an important step in the clinical assessment of left ventricular function using echocardiography. Current guidelines recommend the use of the biplane method, which is known to significantly underestimate cavity volumes. We explored the use of the Efficient Pix2Vox network to reconstruct 3D geometries from standard 2D echocardiographic views, from which volumes could be directly computed. Reconstruction performance was evaluated and compared to data derived from cardiac magnetic resonance (CMR) imaging. The DICE score was 0.856 ± 0.043 for end-diastolic geometries and 0.800 ± 0.052 for end-systolic geometries. Using CMR as the reference, the intraclass correlation coefficient for the proposed 3D reconstruction method compared to the conventional biplane method was greater for end-diastolic volume (0.860 [0.69, 0.94] versus 0.836 [0.21, 0.95], respectively), but not for end-systolic volume nor ejection fraction. Furthermore, reductions in absolute and proportional biases were observed for both end-diastolic and end-systolic volumes. This study demonstrates the potential of reconstructing 3D cardiac geometries from 2D echocardiograms for more reliable volumetric assessment.