Chapter 3 - Anatomical photo representations for cardiac imaging training

Abstract

Advances in cardiac imaging techniques play a vital role in the fight against cardiovascular disease by helping to achieve early and accurate diagnosis and guidance in cardiac intervention. In addition to these imaging systems, computer-aided diagnostics (CADs) are essential in modern healthcare to provide useful and comprehensive clinical information in a short period of time. However, due to the technical limitations imposed by the imaging systems, CAD systems have not been widely adopted in clinical practice. Cardiac cinema magnetic resonance (MR) imagery, in particular, has low through-plane resolution and may contain motion artifacts due to the acquisition process. Similarly, the quality of cardiac ultrasound images depends on the operator and often has a poor signal-to-noise ratio, which presents difficulties for automated image analysis. To address these limitations and enhance the accuracy and robustness of automated image analysis, this thesis focuses on the development and application of state-of-the-art machine learning (ML) techniques in the field of cardiac multifaceted learning. Specifically, we propose new types of representation of image features that are learned from ML models and aimed at highlighting correspondence between multimodal images. Such depictions are often intended to depict the cardiac anatomy in more detail for better understanding and study. In addition, we discuss how quantitative research can benefit from the use of these trained image representations in segmentation, motion-tracking, and multimodal image registration. Specifically, a probabilistic edge-map representation is implemented to define anatomical correspondence in multimodal cardiac images and to demonstrate its use in spatial image alignment and anatomical localization. In addition, a novel image super-resolution system is implemented to improve cardiac cinema MR images and we demonstrate that high-resolution image representation can be useful and informative for various types of subsequent research, including volumetric measurements and strain analysis.