Currently, MEG helmets with over 200 sensors are already used to detect magnetic brain fields by means of a sensitive transducer technology called Superconducting Quantum Interference Device (SQUID). This provides the doctor with a host of data offering the finest temporal and highest spatial resolution. Specialists can detect a disorder by observing the complex brain wave form and analysing the frequency content. The doctor has to separate the MEG data into signal classes. Each of these contains a certain frequency band which allows to localise the signal's source. To this end, wavelet cross-correlation analysis is used. Unlike the traditional Fourier-based analysis, wavelet-based analysis has the capability to explore the frequency content without losing the time information of the original brain data.
Though brain data analysis is mostly performed on a single processor basis, the development of optimised measurement technologies has increased the amount of data collected from imaging modalities like Magnetic Resonance and Computed Tomography, or from brain signal capturing techniques like electroencephalography (EEG) and MEG. As a result, the time duration for analysis has risen as well, albeit that recent processor technologies enable the scientists to perform fast analysing computation. MEG data analysis however is very computation-intensive and time-consuming.
Since wavelet cross-correlation analysis requires both coarse-grained and fine-grained parallelism, the Japanese team had to resort to parallel and distributed computing using the Globus Grid Toolkit and MPICH-G, which is a Globus implementation of Message Passing Interface (MPI), a specification of the communication library for parallel programming on a distributed memory system. The Globus Grid toolkit allows to perform global-scale distributed parallel computing, since it is capable of cancelling the difference among administrative domains, even across national borders. Therefore, it has the potential to offer a wide range of powerful computational resources.
The upcoming of advanced network technologies such as IPv6, allows MEG analysis data to be transferred in the order of one Gigabit in a few seconds, which will be increased to even a TeraByte per second in a couple of years. This enables physicians to analyse MEG data as quickly as possible, in order to timely detect early symptoms of possible brain disease. The sending of MEG data over the Internet requires the guarantee of patient privacy and security. Therefore, the MEG project team utilises advanced cryptography and network technologies, such as Public Key Infrastructure (PKI), IP Security (IPSec), and Secure Shell (SSH).
Further advantages of a geographically and organisationally distributed MEG system are that the medical, computational and signal processing knowledge can be shared easily among the partnering experts. There are only a few tens of MEG installations across the world, necessitating the distribution of resources. The examination room has to be shielded against environmental noise and liquid helium to fully realise the superconducting effect. MEG is very sensitive to environmental noise and risks to capture external noises as well as brain signals. All this makes the purchase and maintenance of MEG equipment very expensive because of its advanced noise cancelling features.
In addition to the parallel computing and data acquisition components, the MEG project also includes plans for a Graphical User Interface (GUI) and an intuitive method for data visualisation to show the source of a specific brain signal to the physician. This will be done by superimposing the results of the wavelet and independent component analyses on human anatomical 3D images that are acquired from real patient Magnetic Resonance Imaging and Computed Tomography data. This might eventually open the door for use of real time visualisation of MEG data in computational surgery. More technical details about wavelet and ICA-based signal processing are available at the MEG project Web site.