In the industrialised countries, cardiovascular diseases constitute the first life-threatening illness among the population. Exact diagnosis and proper treatment include accurate manual or computational measurement of the patient's heart geometry. The 3D HeartView project, funded under the European ESPRIT programme, introduces innovative software for the creation of a 3D heart model out from a sequence of 2D X-ray angiographic projections. The system makes use of High Performance Computing (HPC) on parallel platforms in order to reduce the modelling time to less than ten minutes, which is considered acceptable by clinical standards. The practical project aims are to have the software validated by the physicians of the partner clinics and to convince the industrial key-players of the upgrading value this new technology has to offer to their existing equipment.
The reconstruction of a 3D heart model is performed by following a number of well defined steps. The patient first gets an injection of contrast agent in the heart chambers, ventricles or vessels. Simultaneously, the gantry is rotated around the patient in order to obtain several images for direct or indirect digitisation. All modern angiographic devices normally provide a DICOM-3 (Digital Imaging and Communications in Medicine) digital image output. Otherwise, a frame grabber can digitise the video image stream in real-time. A synchronously running ECG records the cardiac phase, associated with each single image. Since different view angles are produced, it is important to match only images of the same cardiac time-step for the final reconstruction. As a result, you get a volumetric dataset, to be used for further volume rendering procedures.
To minimise the modelling time, the existing software is ported to a Parsytec CC-i system with four processing nodes. The modelling algorithm has to be parallelised in two ways, namely Message Passing parallelisation as to reflect the distributed memory architecture of the used HPC system, and Shared Variable parallelisation to fully exploit the system's potential. Both the digital images and the corresponding heart phase are introduced into the HPC computer for the generation of a backprojection. As a result, a 3D model of the concerned organ is being produced. The project partners already have anticipated a 4D reconstruction of a pulsating 3D model of a heart ventricle or vessel by means of spatial extension of the heart in all cardiac phases. Unfortunately, it can't be used yet for everyday clinical practice.
Once the reconstruction finished, the volumetric dataset is ready for real-time, interactive diagnostic viewing in a volume visualiser. It is even possible to set in segmentation modules to measure, for instance, chamber volumes or stenosis diameters. The higher the accuracy of the reconstruction, the better unusual pathological deformities are able to be recognised by the cardiologist. This constitutes an important factor for the system's final validation. The great benefit of the fully automatic software implies the liberation of the physician from the tiresome, manual measurement. The diagnostic process will be largely improved in terms of accuracy, objectivity, reliability and repeatability. For a complete account of the technical details, please consult the 3D HeartView project's Web site.