< Back to previous page


Model-based image computing for multi-X medical image data.

Challenge. Medical imaging today plays a crucial role in all stages of the medical decision process, not only for early patient diagnosis and individualized therapy planning, but also for population screening, therapy outcome prediction and assessment, and also in translational pre-clinical and clinical research. Tomographic imaging modalities, such as computed tomography (CT), magnetic resonance imaging (MRI) or positron emission tomography (PET), allow acquiring three-dimensional (3-D) images of the interior of the body in a minimally invasive way. Different modalities are based on different physical properties and therefore provide complementary morphological, functional and molecular information. However, recent innovations in medical imaging technology, such as multi-slice (volumetric) and multi-energy CT, multi-spectral and multi-frame MRI, multi-modal and multi-tracer PET/CT, multi-array and multi-dimensional US, or multi-planar and multi-contrast interventional imaging, have created a tsunami of imaging data, which is becoming a bottleneck in radiological practice. At the same time, the rapid adoption of digital picture archiving and communication systems (PACS) in the radiological workflow makes that large databases of clinically documented images are becoming available. These databases create new opportunities for multi-modal, multi-temporal, and multi-subject assessment, such as detecting changes over time in a single individual by comparing baseline and follow-up images or detecting pathological abnormalities by comparing patient images with images of control subjects. In order to optimally exploit all available imaging data in such a multi-X imaging environment, reliable computer-aided image analysis becomes indispensable to extract and quantify the relevant information from the multi-X image data, to fuse complementary information and to support the interpretation thereof. Image computing facilitates the quantitative analysis of imaging data through a combination of image processing techniques such as image reconstruction, registration and segmentation. The analysis of multi-X imaging data is complicated by ambiguity induced by the intrinsic limitations of the image acquisition process (limited resolution, noise, artifacts), by the complex shape and intensity appearance of the anatomical objects in the image scene and by the multi-X nature of the data itself. Problem statement. Due to the complexity of medical imaging data and the ambiguity inherent to limitations of the image acquisition process, a prerequisite for medical image computing is the availability of suitable parametric models that incorporate prior knowledge about the typical appearance of the object of interest in the image data. In medical applications these models need to be sufficiently flexible to account for image appearance variations, such as normal biological shape variability and pathological abnormalities. Current flexible geometric models are represented by either a deformable shape (e.g., set of landmarks, analytic curve, tetrahedral grid) or a deformable picture (e.g., atlas or gray value image itself) with geometric and photometric properties. The model properties can have a global or local nature and can be tissue-dependent. They can be specified deterministically, or as statistical distributions obtained from a representative ensemble of images during a training phase. Typically the selection between or combination of these different options is made ad hoc and the model parameters are tuned heuristically for each specific application. This negatively affects the accuracy, the reliability and practical usefulness of current image computing methods. Objective. Our ultimate goal is to develop a unified hybrid model representation without using heuristics. Together with a suitable similarity measure and model fitting method our problem-solving strategy will offer a unified approach to problems of image formation, image fusion and image quantification. We expect (1) that this approach will be applicable to a wide variety of clinical and biomedical applications and (2) that it will improve the accuracy and reliability without the need of fine-tuning and other interactive tasks.
Date:1 Jan 2011  →  31 Dec 2017
Keywords:Image reconstruction, Image segmentation, Medical imaging, Image analysis, Image registration