Segmentation automatique par apprentissage profond des artères coronaires à partir d’angiographie 2D

Abstract

Accurate automatic segmentation of coronary arteries plays a crucial role in medical imaging for the diagnosis, analysis, prevention and treatment of cardiovascular pathologies. In this master’s thesis we propose an innovative approach based on the U-Net convolutional network to segment coronary arteries from 2D angiographic images. To optimize the performance of our segmentation, we have integrated the filters proposed by Meijering et al. and Sato et al. specifically designed for the extraction of tubular structure features. Our experimental approach is based on two distinct stages. To achieve optimal segmentation, we decided to incorporate the Meijering and Sato filters. This approach aims to enhance vessel visibility and improve the distinction between coronary arteries and surrounding tissue. The incorporation of these filters significantly improved results compared with those obtained by U-Net without filtering. In a second phase, gamma correction was incorporated into our approach to assess its impact on enhancing the results. The results obtained were comprehensively assessed using evaluation metrics such as accuracy, sensitivity, specificity and Dice coefficient. This study contributes to the advancement of coronary artery segmentation techniques, paving the way for more reliable and accurate clinical applications. The methods presented in this thesis offer promising prospects for the continuous improvement of cardiovascular diagnostics and the personalization of medical treatments.

Date	23 Feb 2024
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Luc Duong (Supervisor) & Faten M'Hiri (Co-supervisor)