Feasibility of whole heart 3D shape reconstruction from sparse echocardiographic views using CT-derived simulations

To verify the feasibility of reconstructing whole heart from 2D ultrasound sparse views. This study introduces a series of models for whole-heart reconstruction and conducts a benchmark study to evaluate the impact of pose information on reconstruction accuracy using synthetically generated 2D ultrasound sparse views from CT volumes, with the CT volumes serve as the label. Experiments demonstrated that 3D cardiac reconstruction from only 2D views had limited accuracy (mean Dice similarity coefficient (DSC): $$88.86\%$$). With precise pose information, four standard echocardiographic views achieved high precision with a mean DSC of $$95.67\%$$. Even non-standard views combined with imprecise pose information significantly outperformed methods lacking pose information (mean DSC: $$92.33\%$$). Estimation of left ventricular volume further confirmed the superior reliability of reconstructions incorporating pose information. Reconstructing 3D heart from 2D ultrasound views is feasible. Pose information critically improves the accuracy of 3D reconstruction to support downstream clinical tasks, particularly for pathological cases. Acquiring echocardiographic view pose information represents a promising approach to realize personalized 3D cardiac shapes for enhanced clinical diagnosis.