Description
Date depot: 3 septembre 2020
Titre: Machine learning in automated analysis of human aorta using innovative 4D flow MRI
Directrice de thèse:
Nadjia KACHENOURA (LIB)
Encadrant :
Thomas DIETENBECK (LIB)
Domaine scientifique: Sciences pour l'ingénieur
Thématique CNRS : Intelligence artificielle
Resumé: Study context. The aorta is the main artery of the human body and plays a key role in the cardiovascular system as it conveys oxygenated blood to our organs while ensuring constant and continuous blood flow and regulated arterial pressures despite the pulsatile and recurrent nature of the heart ejection. This cushioning function of the aorta can however be impaired by various diseases which stiffen the aorta thus reducing its elastic efficiency. This aortic stiffening is often characterized by structural and mechanical changes of the aortic wall interplaying with circulating blood flow disorganization and is known to play a significant role in cardiovascular disease development and progression towards fatal events (aortic wall rupture).Details of the proposal. A Imaging context and lack in state of the art. Magnetic Resonance Imaging (MRI) is a non-invasive and non-irradiating imaging modality able to provide both aortic anatomy (angiography data) and blood flow dynamics (4D flow data) through volumetric acquisitions. Anatomic angiography data are acquired on a single phase of the cardiac cycle but have a high spatial resolution and contrast. They are usually used in clinical routine to extract morphological indices (diameters on selected slices) and to visualize geometrical malformation (stenosis, dilation, abnormal valves). 4D flow datasets consist in the acquisition of circulating blood flow X, Y and Z velocity components in all aortic voxels during the entire cardiac cycle, but with a lower resolution and overall contrast. In the clinical routine, these data can be visualized but the lack of segmentation tools hinders the extraction of quantitative measures, although they will be of major usefulness for the patient. In our laboratory (Laboratory of Biomedical Imaging, Cardiovascular Imaging team), we have recently developed a semi-automated technique (MIMOSA software) which is able to reliably detect 3D aortic borders on angiographic images (1). This technique has been applied on the data of 400 patients including: controls with a wide age range (20 to 80 years), patients with aortic dilation, and female patients with abnormally small aorta because of their Turner syndrome (clinical data are made available through our tight collaboration with Pitié-Salpêtrière Hospital). To summarize: 1) there is an urgent need for automated segmentation of the above mentioned MRI images to accurately quantify aortic morphology, function and well as hemodynamics (2,3)
Résumé dans une autre langue: .
Doctorant.e: Guo Jia