Detailed 3D reconstruction of a patient's heart geometry allows the physician to accurately diagnose any serious failure within the cardiac structure. At present, the production of a complete 3D data set takes the cardiologist about 1 to 2 hours of manual measurement. The objective of the Esprit funded 3D HeartView project is to implement an advanced method of 3D image processing based on 2D angiography sequences with minimal user intervention. In order to speed up the procedure for routine clinical use, the system is ported on a high performance parallel hardware platform to form a powerful add-on for the digital angiographer. The project partners are eager to prove to the major suppliers of angiography devices that the 3D HeartView accelerator offers valuable improvement for medical problems and as such is bound to open new markets.
The 3D reconstruction technique is based on pyramid-beam backprojection, a method that is also used in computed tomography (CT) and which has been optimized by the Computer Graphics Centre (ZGDV) in Darmstadt. All images corresponding to an identical cardiac phase are selected from the sequence and pre-processed through size equalization. The applied algorithm projects them back into a 3D image space as stacked pixel slices. Automatic contour tracing is used to calculate the object's volume. In this way, the physician obtains a volumetric dataset. Since this kind of backprojection is very time-consuming, project co-ordinator Panos Skoutas SA has parallelized the entire process on a Parsytec CC-i-3D 4x2 parallel machine, in collaboration with ICCS, the Institute of Communication and Computer Systems at the National Technical University of Athens.
Cardiologists of three hospitals in Athens and Münster, all of them with a varying user profile, participate in the project to assess the system prototypes that have been adapted to their specific needs. The reconstruction procedure includes six steps. Initially, the heart chambers or great vessels are injected with contrast agent while a mono- or biplane angiographer is rotated around the patient to acquire X-ray images from different angles. If the device has a DICOM-3 digital output, the images, together with the simultaneously running electrocardiogram (ECG), are automatically stored on the high performance computing (HPC) system via the Digital Imaging and Communications in Medicine (DICOM) image exchange standard. If not, the images have to be digitized in real time at video frame rate with use of the HPC machine's frame grabber. Everything is now in place to start the actual backprojection.
In step four, the HPC computer selects the main cardiac phases, consisting in the end of the diastole and systole period in which the ventricles display their maximal and minimal volume. The partners have extended the system to reconstruct all intermediate stages as well in order to offer the physician a 4D wall motion study to view the pulsating heart from any possible angle but the insufficient performance makes it yet unsuitable for clinical use. The ZGDV partner has already designed the InViVo interactive volume visualizer to perform an accurate diagnosis based on the high resolution reconstructed chamber or vessel section volume of the patient. As a final step, the doctor can automatically have the chamber volume calculated through one simple mouse click inside the cavity. The method requires no special knowledge of computer graphics and can be executed by a medical assistant.
From the industrial side, the 3D HeartView project is supported by Philips, Siemens and Toshiba. These three leading manufacturers of angiography devices together represent a global market share of 70%. The 3D accelerator offers them a chance to upgrade their products and cut down the modelling time to less than three minutes for cases with a medium load. Therapy for patients suffering from cardiovascular diseases will become more adequate thanks to the optimized diagnostic method and HPC facilities for accurate measurement of the heart geometry. For a detailed technical account, we refer to the home page of the 3D HeartView project.