Medical image analysis for the detection, extraction and modelling of vascular structures

Resumen

Vascular-related diseases are among the most important public health problems indeveloped countries. Recent advances in medical imaging provide high resolution 3Dimages of the vessels, allowing the generation of accurate patient-specific geometricvessel models. Image-based vessel analysis allows advanced computer-assisted diagnostic,intervention and follow-up of vascular-related diseases. It also provides valuableinformation for computer-assisted surgery planning and navigation, both to avoiddamaging vital structures and to use vessels as anatomical landmarks.From the modelling point of view, this thesis proposes a Vessel Knowledge Representation(VKR) model in the area of blood vessel analysis. It allows reusability ofsoftware pieces through appropriate abstractions, facilitating the development of innovativemethods, procedures and applications. The VKR model is designed for an easyintegration with existing medical imaging and visualization software platforms.Regarding 3D vascular detection algorithms in medical imaging, we provide a detailedanalysis of some well-known vesselness functions. We identify different typesof scaling parameters in these detectors, and tests their individual influence and theirrelationship against synthetic and real datasets in order to establish some scale selectioncriteria. With respect to 3D vascular extraction methods on angiographic images,we propose an architecture and process model for the subset of vascular tracking methodscalled Generalized Vascular Tracking (GVT). We demonstrate how the differentcomponents and stages of the GVT model allow incorporating different modules intothe system with increasing complexity. We also contribute with a novel method foroptimized vascular section estimation during vessel tracking procedures.As a real life application and a particular case of vascular analysis we providea novel segmentation method for both the lumen and thrombus of abdominal aorticaneurysms on CTA images after endovascular intervention and a method for the automaticdetection and quantification of endoleaks in the thrombus.Finally, we developed the Image-based Vascular Analysis (IVAN) Toolkit, a set ofsoftware libraries for vascular detection, analysis and modelling in medical imaging. Itimplements many of the ideas proposed in this Thesis, and has been used and validatedextensively in our experiments.