Fully automatic assessment for left ventricle image segmentation and feature extraction

• Autor(a) principal: Neto, António Maria Sousa Gomes
• Data de Publicação: 2023
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10071/31155
• topic: Deep learning; Machine learning; Left ventricle; Medical image segmentation; Feature extraction; Cardiac MRI; Ventrículo esquerdo; Segmentação de imagem médica; Extração de características; Ressonância magnética

The segmentation and measurement of the left ventricle (LV) and the myocardium (MYO) are common clinical practices in the diagnosis of left ventricular conditions. Magnetic resonance (MRI) is recognized as a reference for non-invasive assessment of left ventricular function. Many studies over the last decades have focused on improving automatic segmentation and classification approaches, with multiple datasets for cardiac MRI. However, most automatic methods fail to provide reasoning and clinically relevant data to support the diagnosis. In this dissertation, we propose a novel fully automatic pipeline for left-ventricular segmentation and function assessment. The proposed pipeline leverages a state-of-the-art segmentation method (nnU-Net), which is then enhanced through the application of two active contours for smoother and more robust contour delineation. Automatic feature extraction methods are then proposed for the assessment of both local and global features within cardiac structures. Results show that the presented segmentation enhancements approach maintains accuracy while improving feature-extraction capabilities and robustness. Furthermore, the proposed feature extraction methods proved effective in providing clinically relevant data for assessing left ventricular function.