CardioMorphNet: Cardiac motion prediction using a shape-guided Bayesian recurrent deep network

Area of Science:

• Medical Imaging
• Cardiovascular Imaging
• Artificial Intelligence in Medicine

Background:

• Accurate cardiac motion estimation from cine cardiac magnetic resonance (CMR) images is crucial for assessing cardiac function and detecting abnormalities.
• Current methods often fail to capture heart motion accurately due to reliance on intensity-based registration, which can overlook critical anatomical regions.

Purpose of the Study:

• To introduce CardioMorphNet, a novel recurrent Bayesian deep learning framework for 3D cardiac shape-guided deformable registration using short-axis (SAX) CMR images.
• To improve the accuracy and reliability of cardiac motion estimation and clinical index calculation in cardiovascular imaging.

Main Methods:

• CardioMorphNet utilizes a recurrent variational autoencoder to model spatio-temporal dependencies across the cardiac cycle.
• It employs Bayesian posterior models for bi-ventricular segmentation and motion estimation, guiding registration via segmentation maps to focus on anatomical regions.
• The framework leverages sequential SAX volumes and spatio-temporal features, avoiding intensity-based similarity losses and enabling uncertainty quantification.

Main Results:

• CardioMorphNet demonstrated superior performance in cardiac motion estimation compared to state-of-the-art methods on UK Biobank and M&M datasets.
• Uncertainty assessment revealed lower uncertainty values in the cardiac region, indicating higher confidence in its motion field predictions.
• Clinical index extraction assessment showed CardioMorphNet estimates these indices more accurately than other approaches.

Conclusions:

• CardioMorphNet offers a robust and accurate method for cardiac motion estimation from SAX CMR images.
• Its shape-guided Bayesian deep learning approach enhances precision and provides reliable uncertainty quantification for clinical applications.
• The framework shows significant potential for improving the diagnosis and management of cardiac conditions through more accurate functional assessments.