Cardiovascular diseases are the leading cause of global mortality and a major contributor to disability worldwide, and clinical management of valvular heart disease has become a major focus of cardiovascular medicine because of its growing prevalence. Aortic valve abnormalities are among the most common types of valvular heart disease that are associated with aging and congenital or chronic heart diseases. However, their surgical treatment requires a thorough morphological comprehension of the complex three-dimensional anatomy of the aortic root and valve, which can be preoperatively assessed by image-assisted cardiovascular examinations and proper interpretation of quantitative features extracted from the acquired images. In the past few decades, advances in artificial intelligence, especially deep learning, have become an integral part of state-of-the-art automated computer-assisted medical image analysis, with a large spectrum of applications. In the proposed research project, we will design, develop and evaluate deep learning-based image analysis that aims to improve medical image interpretation, and its merging with clinical practice in the field of cardiovascular medicine and cardiovascular imaging, in particular for the clinical management of valvular heart disease. In cardiovascular medicine, automated computer-assisted approaches for aortic valve morphometry are mostly concentrated around image segmentation, the process of delineating the aortic root boundaries in the given image, and landmark detection, the process of identifying specific anatomical points that define the aortic valve and its cusps. To achieve a comprehensive framework for aortic valve morphometry, which is the quantitative analysis of form that is described by features of size and shape, we will first devise a database of cardiac three-dimensional computed tomography images with corresponding reference landmark annotations. We will then design and develop state-of-the-art deep learning algorithms for aortic valve landmark detection and aortic root image segmentation, with the focus on reinforcement learning, an emerging field within artificial intelligence that is defined as the science of decision-making. By reformulating image analysis tasks as a collaborative multi-agent reinforcement learning method, the agents will dynamically learn the optimal policies based on the information provided by the other agents. We will then develop and apply medical image analysis techniques for automated measurement of morphometric features that describe the complex three-dimensional anatomy of aortic cusps. These features are of significant importance for cardiovascular surgeons in the process of planning aortic valve surgical replacement or repair, and their automated measurement has, besides being repeatable and reliable, the potential to facilitate the workflow of our clinical partners and, ultimately, improve the quality of life of patients with valvular heart disease. Finally, we will devise a software framework to merge the developed deep learning framework and aortic valve morphometry with tools for medical image manipulation, and ensure their end-to-end functionality. The methodological advances and the obtained statistical analysis of the results will be disseminated by publication in top-ranking scientific journals and presentation at international conferences, as well as by means of open science and networking. Our objectives are therefore concentrated around a comprehensive framework for a complete aortic valve morphometry that will be based on methods for aortic valve landmark detection and aortic root image segmentation based on state-of-the-art advances in deep learning, integrated in an end-to-end functionality software that will facilitate the clinical workflow from the perspective of cardiovascular image-based examinations and surgery planning, and enable further research in the field of cardiovascular medicine.
