Automatic extraction of 3D dynamic left ventricle model from 2D rotational angiocardiogram

Area of Science:

• Medical Imaging
• Computational Cardiology
• Biomedical Engineering

Background:

• Accurate 3D modeling of the left ventricle (LV) is crucial for assessing cardiac function and guiding interventions.
• Existing methods often require complex imaging or manual segmentation, limiting their clinical applicability.
• Sparse 2D rotational angiocardiograms present challenges due to limited data and motion artifacts.

Purpose of the Study:

• To develop an automated method for direct extraction of 3D dynamic LV models from sparse 2D rotational angiocardiograms.
• To enable quantitative cardiac function analysis using the extracted dynamic LV models.
• To provide visual guidance for cardiac interventions by overlaying the 3D model onto real-time fluoroscopic images.

Main Methods:

• Utilized ungated CT reconstruction for initial static LV model extraction, despite motion artifacts.
• Projected the static LV model onto 2D angiocardiogram images and deformed its silhouette to match the LV blood pool boundary.
• Back-projected silhouette deformation vectors to 3D space as anchor points for Thin Plate Spline (TPS) interpolation of the dynamic 3D mesh.

Main Results:

• Validated the method on 12 synthesized datasets, achieving a mean point-to-mesh error of 0.51 ± 0.11 mm, indicating high accuracy.
• Demonstrated promising preliminary results on two real-world datasets, including patient and animal (pig) data.
• The extracted 3D LV meshes closely matched the ground truth in synthesized data.

Conclusions:

• The proposed automated method effectively extracts accurate 3D dynamic LV models from sparse 2D rotational angiocardiograms.
• The method shows potential for quantitative cardiac function analysis and real-time guidance in cardiac interventions.
• Further validation on larger and diverse clinical datasets is warranted to confirm its clinical utility.