Automatic segmentation of the left ventricle in cardiac magnetic ressonance exams

• Grant number: 19/22116-7
• Support type: Scholarships in Brazil - Master
• Effective date (Start): March 01, 2020
• Effective date (End): September 30, 2021
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal researcher: Fátima de Lourdes dos Santos Nunes Marques
• Grantee: Matheus Alberto de Oliveira Ribeiro
• Home Institution: Escola de Artes, Ciências e Humanidades (EACH). Universidade de São Paulo (USP). São Paulo , SP, Brazil
• Associated research grant: 14/50889-7 - National Institute of Science and Technology Medicine Assisted by Scientific Computing (INCT-MACC), AP.TEM

Abstract

Cardiac magnetic resonance imaging is recognized as the gold standard for the diagnosis of various heart diseases. From the segmentation and analysis of the left ventricle in these exams, clinical metrics used in diagnoses can be obtained. However, manual segmentation of the left ventricle in the numerous magnetic resonance images requires time and repetitive effort from the expert, which may increase the diagnostic quality variability. In the literature, many methods have been proposed for automatic segmentation of the left ventricle. The main approaches use image-based methods, shape prior and artificial intelligence. Despite obtaining good results, no method has yet achieved expert excellence due to the wide variation of the structures represented in magnetic resonance images. From a systematic mapping, we found that the use of hybrid methods based on shape prior and artificial intelligence have achieved promising results and offer a possible solution to the segmentation problem. The aim of this project is to develop a new hybrid method for automatically segmentating the left ventricle based on artificial intelligence and shape prior. The method aims to reduce the occurrences of anatomically impossible segmentations, one of the greatest drawbacks of the state-of-art methods, without compromising the quality of segmentation. Results will be validated by measuring classic segmentation evaluation metrics as well as clinical metrics used in the diagnosis. Besides offering a contribution to the computer graphics area, by proposing a new segmentation method, the project contributes to the area of computer-aided diagnosis in Cardiology. (AU)