RVDeformer: Sparse Point Cloud-Guided Right Ventricle 3-D Reconstruction in Echocardiograms

Area of Science:

• Medical Imaging
• Computational Anatomy
• Cardiovascular Engineering

Background:

• Accurate 3D reconstruction of the Right Ventricle (RV) is vital for assessing cardiac function.
• Existing 3D reconstruction methods struggle with complex RV anatomy and limited data from 2D echocardiograms.

Purpose of the Study:

• To develop a novel framework, RVDeformer, for precise 3D RV reconstruction from sparse point cloud data.
• To enhance the clinical evaluation of cardiac function through improved RV imaging.

Main Methods:

• Proposed RVDeformer, a sparse point cloud-guided framework that treats reconstruction as a mesh deformation problem.
• Utilized an end-to-end neural network (RVDeformNet) and a point cloud-mesh fusion module for feature extraction and vertex displacement prediction.
• Trained and validated the model on a large clinical dataset of 1,278 cases.

Main Results:

• RVDeformer achieved superior performance compared to state-of-the-art methods.
• Quantitative results include a Chamfer Distance (CD) of 2.24±0.55 mm, F1-score of 0.74±0.10 (3 mm threshold), and Volumetric Similarity (VS) of 91.53±2.28%.
• Demonstrated significant potential for clinical applications in cardiac imaging.

Conclusions:

• RVDeformer offers a robust and accurate solution for 3D RV reconstruction from echocardiograms.
• The framework effectively addresses the challenges posed by complex anatomy and incomplete spatial information.
• The method shows promise for advancing clinical assessment of cardiac function.