CardioSynth: Parameter-driven cardiac MRI generation via oriented bounding boxes.

Area of Science:

• Medical Imaging
• Computational Anatomy
• Artificial Intelligence in Healthcare

Background:

• Cardiac magnetic resonance imaging (MRI) offers detailed anatomical insights but is expensive and impractical for routine monitoring.
• Precise control over cardiac substructure areas is crucial for understanding cardiac development, adaptation, and disease progression.

Purpose of the Study:

• To develop a novel framework for synthetic cardiac imaging with parameter-driven control over substructure areas.
• To enable anatomically plausible modifications of cardiac substructures for research and monitoring.

Main Methods:

• A three-stage framework integrating oriented bounding box (OBB) encoding of cardiac substructures.
• Progressive label modification algorithms for parameter-driven area changes while ensuring anatomical consistency.
• Diffusion-based models for generating realistic cardiac MRI from bounding-box representations and segmentation masks.

Main Results:

• OBB encoding provides superior control over cardiac substructure area modifications compared to traditional methods.
• Systematic deviations of approximately 5% in area modification were corrected to within ±1% using calibrated inputs.
• Synthesized cardiac MRI images demonstrated anatomical plausibility and structural consistency.

Conclusions:

• The proposed framework facilitates parameter-driven synthetic cardiac MRI generation with accurate substructure area control.
• Combining OBB encoding, progressive modification, and diffusion modeling enhances anatomical consistency.
• This approach supports longitudinal monitoring and disease progression studies by reducing the need for repeated scans.