Core Concepts
An intensity-based optimization algorithm that simultaneously aligns all short-axis and long-axis cardiac MRI slices by maximizing the pair-wise intensity agreement between their intersections, without requiring any prior anatomical knowledge.
Abstract
The content describes a method for correcting misalignments between cardiac MRI slices acquired during different breath-holds. The key points are:
Cardiac MRI imaging typically involves acquiring short-axis (SA) and long-axis (LA) slices, which can become misaligned due to variations in the heart's location across acquisitions.
The proposed approach formulates the alignment problem as a subject-specific optimization that jointly optimizes the 3D rotation and translation parameters of all SA and LA slices to maximize the intensity similarity at their intersections.
Unlike previous methods, this approach does not require any prior knowledge about the underlying cardiac anatomy, and solely relies on the image intensity information.
The method is evaluated quantitatively by synthetically misaligning 10 motion-free datasets and demonstrating the algorithm's ability to reliably recover the original alignment parameters, even under large rotation and translation perturbations.
The results show the algorithm performs better at correcting rotations compared to translations, with the most challenging scenario being pure translations.
Future work includes exploring more robust optimization techniques and incorporating intensity-invariant similarity metrics like mutual information.
Stats
Cardiac MRI datasets from 10 subjects in the UK Biobank were used, each containing 9-13 SA slices and corresponding 2-chamber, 3-chamber, and 4-chamber LA slices.