Automating Model Generation for Image-Based Cardiac Flow Simulation.

Area of Science:

• Medical imaging
• Computational fluid dynamics
• Cardiovascular modeling

Background:

• Patient-specific computational fluid dynamics (CFD) modeling of left ventricle (LV) flow aids in functional assessment.
• Current model construction is time-consuming, requiring manual segmentation and separate software tools, limiting large-scale analysis.

Purpose of the Study:

• To develop a fully automated approach for generating CFD-suitable LV models from medical imaging data.
• To significantly reduce the time and human effort involved in LV CFD model generation.

Main Methods:

• Leveraged deep learning-based segmentation (ensemble of 2D CNNs) for automatic cardiac structure segmentation from 3D patient images (MR and CT).
• Integrated geometry processing and image registration techniques for robust CFD-suitable LV mesh reconstruction.
• Validated segmentation performance against state-of-the-art methods.

Main Results:

• The automated segmentation approach outperformed existing techniques on benchmark cardiac scans.
• Successfully generated CFD-suitable LV meshes for 78 out of 80 test cases using segmentation and geometry processing.
• Demonstrated the framework's feasibility for LV hemodynamics modeling through CFD simulations on patient-specific image datasets.

Conclusions:

• The proposed framework enables fully automated, rapid generation of CFD-suitable LV models.
• This automation significantly reduces barriers to large-scale patient-specific cardiovascular flow analysis.
• The approach shows strong potential for advancing clinical applications of cardiac CFD modeling.