Hyperparameter optimisation and validation of registration algorithms for measuring regional ventricular deformation...

Area of Science:

• Cardiovascular Imaging
• Medical Image Analysis
• Computational Biology

Background:

• Retrospective cardiac computed tomography (CCT) enables myocardial function assessment.
• Generic image registration hyperparameters require optimization for CCT motion tracking.
• Limited validation and open-source tools exist for CCT strain analysis.

Purpose of the Study:

• To optimize image registration hyperparameters for tracking cardiac motion in CCT.
• To validate an open-source platform for measuring regional myocardial strain.
• To assess the impact of temporal sampling on cardiac mechanics quantification.

Main Methods:

• Developed an open-source platform using image registration warping fields on a left ventricular endocardial mesh.
• Optimized hyperparameters for two intensity-based registration methods.
• Validated strain calculations against anatomical landmarks and idealized problems.

Main Results:

• Optimized registration achieved a mean error of 6.63% in tracking anatomical measurements.
• Methods differentiated normal from dyssynchronous contraction patterns.
• 10 temporal frames provided sufficient accuracy for cardiac mechanics quantification, with a maximum error of [Formula: see text].

Conclusions:

• Intensity-based registration with optimal hyperparameters accurately quantifies regional myocardial strain from CCT.
• The open-source platform demonstrates sensitivity for identifying heart failure.
• CCT analysis of cardiac mechanics is feasible and accurate with 10 temporal frames.