- Sven Lončarić and Davor Miličić, Project leaders
- Vedrana Baličević
- Hrvoje Kalinić
- Maja Čikeš
- Bart Bijnens, University Pompeu Fabra, Barcelona, Spain
Faculty of Electrical Engineering and Computing, University of Zagreb, Croatia
University Hospital Rebro, University of Zagreb, Croatia
Ongoing research started in 2007.
Analysis of cardiac ultrasound images is an important tool for diagnosis of cardiac diseases. The hypothesis of the research project is that Doppler ultrasound images of aortic outflow velocity profiles have different features for healthy and diseased cases. The goal of the project is to develop a signal and image analysis methodology to extract relevant structures and features from DICOM Doppler ultrasound images to be used for statistical analysis of aortic outflow profiles for a larger number of patients.
|Aortic ouflow ultrasound image||Statistical model (example)|
From segmented image the desired objects can be separated from the background, measured, counted or in other means quantified, teherefore, the image segmentation is often the first step in image analysis. However, issues such as low spatial (or temporal) resolution, ill-defined boundary, poor contrast, acquisition artifact or other noise place additional demands on segmentation. These issues are common for a noninvasive diagnostic methods such as (cardiac) ultrasound, and are usually solved with the aid of a prior knowledge. Usually the source of such prior knowledge is an expert cardiologist.
In the 20th century the amount of work done by the cardiologist increased significantly, and, e.g. Croatia saw an increase of more than 60% in coronary heart disease death rates between 1988 and 1998. In the case of medical applications, the automatic interpretation of images can relieve clinicians from the labor intensive aspects of their work while increasing the accuracy, consistency, and reproducibility of the interpretations, in the meantime allowing them to focus more on other aspect of their work. The idea is to incorporate the knowledge within the segmentation process in the form of the model that will be used as a prototype for segmentation of desired object. Comparative advantage of the model-based segmentation with respect to the other segmentation methods is the ability to segment the image with no well defined relation between regions and pixels’ intensities. This is usually the case when the objects of the same structure need to be segmented (i.e. have the same texture), and the information about difference between these object is incorporated in spatial relationship between them, other objects, or within their morphometric characteristics. By utilizing the registration process all the statistical (empirical) information contained in the model can be used to segment and quantify the objects in the image and, hopefully, aid the diagnostic process.
|Aortic outflow profile before geometric deformation||Aortic outflow profile after geometric deformation|
New methods for construction of image atlases have been developed, as well as methods for atlas-based segmentation. Atlas-based segmentation is based on registration technique utilizing geometric transformation, image similarity measure (metric) and optimization algorithm. Within the scope of this project a geometric transformation specially formulated for this type of images is used, novel image similarity measure is developed and various methods for atlas construction are investigated. The international research project involves researchers at University Hospital Rebro in Zagreb, Croatia and Pompeu Fabra University in Barcelona, Spain.
A complete list of publications can be found here.