Chapter 4 - Cardiac function review by machine learning approaches

Abstract

Hardware limitations imposed by medical imaging systems can hinder the diagnosis and treatment of cardiac-related pathologies. In the last two decades, computer-aided systems for image processing have been developed to promote these procedures and provide improved clinical outcome. This chapter offers a brief description of these theoretical structures and their applications in segmental cardiac imaging for segmentation, image enhancement, and multimodal image alignment. These analytical methods share common goals, such as time efficiency, quantitative objective evaluation, enhancement, and analysis of multimodal image data. In this survey, we concentrate on how the learning of image representation will lead to these goals and improve the accuracy and robustness of the techniques applied. Our main emphasis is especially on anatomical priors in the segmentation of medical images and also on intermediate feature spaces used in multimodal image registration and super-resolution problems. In addition, this chapter addresses computational strategies focused on machine learning (ML), which have recently gained popularity in the medical imaging community with the advent of annotated datasets. We present two widely used models in this regard and address their aspects in terms of design versatility, model parameterization, and functional aspects (e.g., specifications for computation). Later, an overview of ML applications in cardiac imaging is given, which also includes other areas of application such as landmark localization and control of image quality.