CHeart: A Conditional Spatio-Temporal Generative Model for Cardiac Anatomy

Area of Science:

• Medical image analysis
• Computational biology
• Machine learning

Background:

• Assessing cardiac anatomy and motion is crucial in medical imaging.
• Understanding the link between cardiac structure/function and clinical factors (e.g., age, gender, disease) remains a challenge.
• Current methods for modeling these associations are limited.

Purpose of the Study:

• To propose a novel conditional generative model for describing 4D cardiac anatomy.
• To integrate non-imaging clinical factors as conditions within the generative model.
• To investigate the influence of clinical factors on cardiac anatomy and enable realistic sequence generation.

Main Methods:

• Developed a conditional generative model tailored for 4D spatio-temporal cardiac data.
• Integrated clinical factors (gender, age, diseases) as conditional inputs to the model.
• Evaluated model performance on anatomical sequence completion and sequence generation tasks.

Main Results:

• The model achieved high performance in anatomical sequence completion, matching or exceeding state-of-the-art methods.
• The model successfully generated realistic synthetic 4D cardiac anatomies conditioned on clinical factors.
• Generated sequences exhibited distributions similar to real cardiac imaging data.

Conclusions:

• The proposed conditional generative model effectively captures the relationship between clinical factors and cardiac anatomy.
• The model is capable of generating high-fidelity, clinically relevant 4D cardiac image sequences.
• This work advances the potential for personalized cardiac image analysis and simulation.