Personalized learning-based segmentation of thoracic aorta and main branches for diagnosis and treatment planning

Abstract

Coarctation of the aorta (CoA), is an obstruction of the aortic arch present in 5 − 8% of congenital heart diseases. For children older than a year, CoA is increasingly treated by aortic stenting instead of surgical repair. In pediatric cardiology, CMR is accepted as the standard non-invasive imaging modality to assess the aortic arch in it's entire spatial context [1]. Interpreting such 3D datasets are required to assess the underlying anatomy during both diagnosis and therapy planning phases. However this process is time consuming and varies with operator skills. Within this study we propose -for the first time in our knowledge -a method of automatic segmentation of the lumen of thoracic aorta and main branches. The personalized model of the aorta and the supra-aortic arteries, automatically estimated from 3D CMR data, will provide better understanding of the complexity of pathology and assist the cardiologist to choose the best treatment and timing of repair. A hierarchical framework based on robust machinelearning algorithms is proposed to estimate the personalized model parameters. Experiments throughout 212 3D CMR volumes demonstrate model estimation error of 3.24 mm and average computation time of 8 sec. combined with clinical evaluation on 32 patients.

Non-Invasive Hemodynamic Assessment of Aortic Coarctation: Validation with In Vivo Measurements

“…The overall pre-processing pipeline is illustrated in Figure 4. The details of the image segmentation and other geometric pre-processing steps were reported previously 25 , together with a validation study with clinical evaluation.…”

Noninvasive hemodynamic assessment, treatment outcome prediction and follow‐up of aortic coarctation from MR imaging

“…In the interest of easing translation of personalized CoA models into wider clinical practice, we set out to meet certain expectations. On this course, we have broadly extended our previous works [23][24][25] with stent segmentation, post-and virtual stenting (VS) hemodynamic models. Thus, our contribution provides:…”

“…35 Segmentation of the aorta, including the supra-aortic arteries from MRI, was initially presented in our previous work. 24,25 Multiple groups investigated CoA hemodynamics through computational modeling. Recent studies have suggested that good agreement may be reached between measured and simulated hemodynamic and morphologic indices if subjectspecific boundary conditions are employed.…”

3D Geometric Analysis of the Pediatric Aorta in 3D MRA Follow-Up Images with Application to Aortic Coarctation

“…In previous work on vessel segmentation of the aorta and its main branches, different types of approaches have been used. For instance, approaches based on the Hough transform (e.g., [6]), region growing (e.g., [7]), atlas registration (e.g., [8]), level sets (e.g., [9]), statistical shape models (e.g., [10,11]), machine learning techniques (e.g., [12]), and cylindric intensity models (e.g., [13,14]). However, in previous work often only CTA image data have been used (e.g., [10,7,8,11]), aortic pathologies such as stenoses have often not been considered (e.g., [6][7][8][9]11]), or only the abdominal and descending parts of the aorta are segmented (e.g., [10]).…”

“…Fur-thermore, previous segmentation approaches are typically applied to image data from adults whereas work on pediatric image data and follow-up studies can hardly be found. In [12], an approach is proposed for the segmentation of the thoracic aorta in adult and pediatric MRA image data. The approach is based on machine learning techniques using a multi-stage learning and probabilistic boosting-tree.…”