Ulrich Hartmann is working at the NEC Europe C&C Research Laboratories in Sankt Augustin, Germany. During Parco 99, he presented a new finite element method (FEM) based approach on parallel platforms to produce mechanical head models. The head models are used in three areas of interest. The first focuses on discovering the mechanisms behind brain injuries after impact; the second application field relates to neurosurgery planning and educational training, including the estimation of mechanical consequences of tumour growth; and the third one corresponds to the needs of forensic medicine physicians who have to reconstruct accidents in order to present their results before court.
NEC has developed a comprehensive finite element head model which can pay excellent services in a variety of medical problems. The initial challenge consisted in the design of an automatic tool for the generation of high resolution finite element meshes of the human head. The meshes have to reflect the specific neuro-anatomy of the individual. The second task was to create an adequate modelling environment for medical use. In third place, validation is of crucial importance in order to compare the simulation results with real world data, as to define the limits of the model. Most of the current models are two-dimensional and are based on a fixed geometry which makes them unsuitable for detailed brain research.
The NEC head model is based on 3D magnetic resonance imaging data sets since MRI shows the general brain structures with high contrast. In a first phase, the data has to be pre-processed through segmentation, assigning a specific label to each structure. In the second step, meshes of ventricles are generated to carry out simulations. For the NEC head model, two differently acquired data sets are used and matched, the first one showing the internal brain structures while the second one displays the skull density in contrast. The combined data sets serve as input for a mesh generation in three steps. First, image elements are collected from bricks, resulting in a brick mesh. In the regions where the object is homogeneous, bricks are used to build cells. Then, the cells are sequentially being processed to generate a new mesh.
Until now, supercomputers have not been used to generate meshes but there are two substantial reasons to finally start applying their power. The data sets display a complex geometry for which a high resolution is required to model them. In second place, a lot of new techniques are emerging which measure the material properties of the human tissue by a kind of imaging process. In order to exploit this visual information, high resolution meshes are indispensable. The simulations performed by NEC try to establish the relationship between the positive and negative pressures in the specific brain regions after an impact, and possible brain injuries.
At the NEC C&C Research Laboratories, there are three systems available, which are the NEC CENJU 3 with 128 processors, the NEC CENJU 4 with 64 processors, and a PC cluster with 32 processors. Experiments have proven that the CENJU 4 is the fastest machine by a factor of 3 or 4. The PC cluster is able to compete with the CENJU 3 in performance, which means that PC clusters are becoming a challenge for midrange supercomputers. The total dynamic analysis is comprised in 50 time steps, taking 20 minutes on the CENJU 4. This might be further decreased by more advanced non-linear techniques.
In order to check whether the simulations meet reality, validation is needed. Therefore, cadaver experiments from the late seventies where cadavers were subjected to impacts on the head in order to monitor the pressure change in the brain, have been used to compare them with computer simulation. The results showed a relatively good agreement in the area of overpressure but in the opposite area, there was a mismatch, that may be due to false boundary conditions. In any case, the technology is far from being perfect and still a lot remains to be done, according to Ulrich Hartmann. A second approach for validation has been proposed by Willinger in 1995. Some volunteers were selected in order to measure the properties of the lowest Eigenfrequencies in their heads. The results were compared to the Eigenfrequencies of the NEC simulation, displaying values and shapes of around 70 to 77 Hz.
At present, the mesh generation tool and the Message Passing Interface or MPI-based modelling environment are in place, as well as a few preliminary validation strategies. What needs to be developed still are some methods for dynamic load balancing and accurate validation, as well as reliable material properties. Future plans to meet all these requirements include the DRAMA project to develop a good library for load balancing. Together with the Max Planck Institute in Leipzig, NEC aims to model the simulation processes and compare them with patient data. A third project will involve the development of a new validation approach, together with forensic physicians, in which accidents will be reconstructed in simulations. The results will be matched against the data, derived from the victims of the accidents.
Some novel techniques are coming up which are integrating ultrasound with magnetic resonance imaging. This modality combination yields images of the head with elastic properties. If this can be incorporated within two years into the high performance computing mesh generator, the NEC head model might become an efficient real time application and even serve as a supportive tool to neurosurgery planning. More specific details of the research activities are available at the C&C Research Laboratories home page of NEC Europe Ltd.