MRS is a non-invasive technique applied to measure the chemical content of living tissues by accurately defining the peaks in the spectra. MR spectra have the unique advantage of clearly delineating the human tissue's biochemistry although the metabolic data is hard to decipher. Therefore, the INTERPRET team plans to develop, install, and test an automated, objective system for diagnosis in the MR imaging hospital centres, participating in the project. This will enable radiologists to categorise and grade brain tumours entirely on the basis of MR spectral data, given the fact that many existing hospital MR instruments are able to acquire MR images and spectra during the same examination. The chemical information in a spectrum however is entirely different from the anatomical information provided by most imaging modalities, such as MRI for instance.
As a result, first task of the team, including clinical and industrial partners from Spain, France, the United Kingdom, the Netherlands, and Germany, is to collect a large training set of tumour spectra with unequivocal diagnoses. This database will be subject to password protection but can be accessed by the collaborating centres to enter new data via the Internet in a standardised format or to download and consult the existing data. For the assessment of spectra obtained under the conditions of four different agreed protocols, an automated pattern recognition programme will be developed to classify brain tumours with a statistical measure of confidence using an intuitive interface.
To this purpose, the INTERPRET team has already installed a first prototype of the system with a simple point-and click Graphical User Interface (GUI) at the partnering hospital sites to stimulate feedback from the clinicians at the earliest stage possible. The prototype includes three separate windows. The classification window represents each single spectrum as a point in a multi-dimensional space, allowing the clinician to compare an individual spectrum of unknown class with spectra for which the classification is already known. The image window shows the regions in the brain from which the spectra are acquired and also presents a "nosologic image" of the brain that indicates the classification of the various tissues within and surrounding the tumour. The spectrum window displays the individual spectra in a standard format with mean and standard deviation, as well as the spectra of different tumour types.
The consortium expects the automated INTERPRET decision support tool for brain tumour categorisation will help hospital radiologists to diagnose and grade brain tumours using MRS without submitting the patient to the stress of a brain biopsy. If the prototype shows to be successful, the team hopes to extend the INTERPRET system for other types of tumours and diseases in the longer term. More news and detailed illustrations with regard to the project are available at the Web site of the INTERPRET consortium.