Dr. Ayache and his colleagues are about to upset the tradition of the X-ray imaging technique which has been there for more than a century, in a Copernican revolution of medical Information Technology. The objective of Epidaure (Projet Images, Diagnostic Automatique, RobotiquE) is to develop and design tools for the computer analysis of multidimensional, multimodal medical images. Also the user interaction with medical images, particularly in the context of surgical simulation, is focused to refine both diagnosis and therapy, especially when the therapy can be guided by computed images in video-surgery, interventional radiology, and radiotherapy.
Senior research scientist, Dr. Hervé Delingette, offers an example of the new imaging potential in an anatomical liver segmentation. The image of the patient's liver is partitioned into a number of image point subsets which correspond to meaningful objects. The liver image consists of eight separate functional segments, which have to be identified correctly before moving on to actual surgery. The surgeon can remove about half of the organ and still, it will regenerate itself very quickly, growing back to its original size, but only if the physician is able to strategically target the segments through the use of separate colours in the image.
Thresholding is the most simple segmentation procedure and consists in the extraction of regions in which the image intensity surpasses a certain fixed threshold. The extracted regions serve as data in a complementary task like registration, measurement, movement analysis, or visualisation. A second method relies on deformable models as curves or surfaces evolving in a 3D space to more easily define an anatomical or pathological structure. This procedure is very efficient if the doctor can approximately initialise the model around a region of interest. Yet another approach forms multi-scale analysis based on the construction and use of a theory which enables the analysis of an image at variable levels of detail. Multi-scaling is closely related with the technique of anisotropic diffusion, that smoothes the image while preserving important discontinuities.
To locally extract and modify shapes in the liver's image, Dr. Delingette uses mathematical morphology. Differential operators are applied in 3D images to characterise points, lines or singular surfaces. In the end, the INRIA software generates 3D shapes at those points where lesions appeared on the initial CT scan. Dr. Delingette identifies the segments in the diseased area, to have the computer automatically create a safety zone, indicating the region where the surgeon has to cut. This 3D model in fact constitutes a road map for a liver surgery. Physicians at the Gastro-Intestinal Cancer Institute in Strasbourg have already started using the software in clinical situations. In at least one case where the computer results deviated from the doctor's estimation with regard to a liver segmentation, the computer was right.
The following step in computer modelling aims at simulating not only the geometry of the body, but equally its physical and physiological properties. The research in this area deals with defining geometric and biomechanical models of organs and tissues, as to simulate in real time their deformation, cutting out or suturing. The real time constraints require that images have to be synthesised at a rate of 24 Hz, while the forces exerted on the surgical instruments have to be computed at a rate of some 300 Hz. A lot of energy has been spent on spring-mass models, because they allow relatively simple implementation, and reasonable computation times. Finite element models in turn require a more detailed modelling of the biomechanical properties of soft tissues.
In this regard, the Epidaure medical imaging team supports the Chir project which is precisely concentrated on fundamental research in the modelling of deformable organs, the planning and simulation of robotics procedures, and their safe and real time integration. The Chir team particularly addresses the challenges of heart surgery in collaboration with the surgery équipe of Professor Alain Carpentier at the Hôpital Broussais. For issues that are related to 3D image reconstruction, Chir relies on the results acquired by the Prisme project, whereas the RobotVis team offers assistance in computed 3D dynamic visual perception. Technical details, images and demonstrations are available at the INRIA home pages of Chir, Epidaure, Prisme, and RobotVis.