Search

This thesis brings quantitative methods on complex medical image data closer to clinical routine, and bridges the gap between manual and automated processing. Efficient and effective image segmentation methods are described, and applied to currently insufficiently solved problems that are relevant for medical diagnosis and therapy monitoring. The methods are designed to be suitable for software assistants that meet the requirements imposed by a quantitative analysis in clinical imaging: Applicability regarding the required hardware and acquisition protocols, as well as an effective and intuitive user control, robustness to variations of image quality and also of anatomy and pathology, precision of all derived quantitative measures, sensitivity with respect to relevant changes of the examined structures, and efficiency in order to be applicable to very large data sets on standard hardware. The first part of this thesis describes novel methods that build upon and extend the morphological watershed transform that is known as the key concept for image segmentation in Mathematical Morphology, with applications to brain segmentation in magnetic resonance imaging as well as bone removal in computed tomography and CT angiography. The second part focuses on quantitative image analysis in general, as well as the volumetry of specific neuroanatomical structures in particular. Morphological Volumetry is proposed as a framework for the reliable volumetric quantification of complex three-dimensional structures found in tomographic medical images, combining the extended watershed transform and automated histogram analysis. In cooperation with clinical and industrial partners, the proposed methods were successfully evaluated on a great variety of patient and volunteer images, as well as on software phantoms. The methods have shown potential to improve medical diagnosis and quantitative therapy monitoring and to enhance objectivity in medical image analysis in various instances.

This thesis brings quantitative methods on complex medical image data closer to clinical routine, and bridges the gap between manual and automated processing. Efficient and effective image segmentation methods are described, and applied to currently insufficiently solved problems that are relevant for medical diagnosis and therapy monitoring. The methods are designed to be suitable for software assistants that meet the requirements imposed by a quantitative analysis in clinical imaging: Applicability regarding the required hardware and acquisition protocols, as well as an effective and intuitive user control, robustness to variations of image quality and also of anatomy and pathology, precision of all derived quantitative measures, sensitivity with respect to relevant changes of the examined structures, and efficiency in order to be applicable to very large data sets on standard hardware. The first part of this thesis describes novel methods that build upon and extend the morphological watershed transform that is known as the key concept for image segmentation in Mathematical Morphology, with applications to brain segmentation in magnetic resonance imaging as well as bone removal in computed tomography and CT angiography. The second part focuses on quantitative image analysis in general, as well as the volumetry of specific neuroanatomical structures in particular. Morphological Volumetry is proposed as a framework for the reliable volumetric quantification of complex three-dimensional structures found in tomographic medical images, combining the extended watershed transform and automated histogram analysis. In cooperation with clinical and industrial partners, the proposed methods were successfully evaluated on a great variety of patient and volunteer images, as well as on software phantoms. The methods have shown potential to improve medical diagnosis and quantitative therapy monitoring and to enhance objectivity in medical image analysis in various instances.

This thesis brings quantitative methods on complex medical image data closer to clinical routine, and bridges the gap between manual and automated processing. Efficient and effective image segmentation methods are described, and applied to currently insufficiently solved problems that are relevant for medical diagnosis and therapy monitoring. The methods are designed to be suitable for software assistants that meet the requirements imposed by a quantitative analysis in clinical imaging: Applicability regarding the required hardware and acquisition protocols, as well as an effective and intuitive user control, robustness to variations of image quality and also of anatomy and pathology, precision of all derived quantitative measures, sensitivity with respect to relevant changes of the examined structures, and efficiency in order to be applicable to very large data sets on standard hardware. The first part of this thesis describes novel methods that build upon and extend the morphological watershed transform that is known as the key concept for image segmentation in Mathematical Morphology, with applications to brain segmentation in magnetic resonance imaging as well as bone removal in computed tomography and CT angiography. The second part focuses on quantitative image analysis in general, as well as the volumetry of specific neuroanatomical structures in particular. Morphological Volumetry is proposed as a framework for the reliable volumetric quantification of complex three-dimensional structures found in tomographic medical images, combining the extended watershed transform and automated histogram analysis. In cooperation with clinical and industrial partners, the proposed methods were successfully evaluated on a great variety of patient and volunteer images, as well as on software phantoms. The methods have shown potential to improve medical diagnosis and quantitative therapy monitoring and to enhance objectivity in medical image analysis in various instances.