Patients undergoing diagnostic investigations or medical treatment may have multiple examinations or acquisitions within limited time frame from different cross-sectional modalities, both structural and functional. Within clinical and diagnostic practice it has therefore become more and more important to be able to find correspondances and differences between volumetric image data-sets of the same patient either acquired with the same modality at different stages of the clinical workflow or by means of different imaging procedures. For this purpose advanced techniques for volume integration allow overlapping of data sets on the same display engine for immediate visual comparison.
The aim of the present project is two-fold: first, to develop an automatic or semi-automatic tool for testing existing methods for image integration on different clinical data sets, tuning parameters of the algorithms and measuring the accuracy and clinical relevance of the results; second, to formulate new approaches to image integration, based on regularization and Bayesian methods and compare their performances with the tool previously validated.
Intra- or inter-modality integration of Magnetic Resonance (MR) and Computerized Tomography (CT) volumes will be the first testbed for both activities although image fusion involving functional volumes and high resolution structural slices will be also considered. Finally, medical databases from different Hospitals throughout Europe, utilizing Picture Archiving and Communication System (PACS) and Radiology Information System (RIS) developed by Carestream Health will put part of their clinical data at disposal of the researchers of the project for technical validation of the image processing methods and clinical assessment of the procedure’s reliability.
The first and main objective of our research activity is to formulate and implement a fully automatic procedure for testing a set of algorithms for multimodality and monomodality image integration in medical imaging. More in particular:
- applications will be concerned with high-resolution CT and MR data sets;
- the automation of the procedure will involve both the geometrical parameters utilized in the image registration step and the computational parameters tuning the effectiveness of image fusion;
- clinical validation will involve real slices acquired at different Hospitals utilizing Carestream Health PACS/RIS.
As a further goal, the project will compare this validated set of methods for image registration with two more novel approaches:
- an image fusion technique in which the high frequency component of the image from the high-resolution modality is extracted and added to the low-resolution cross-section;
- a Bayesian approach whereby the information from one modality (or from the acquisition at a particular time frame) is coded in the prior probability density to enhance the reconstruction accuracy for the second modality (or for a successive time frame).
