A computer-based decision support system to assist radiologists in diagnosing and grading brain tumours has been developed by the multi-centre INTERPRET project. Spectra from a database of 1 H single-voxel spectra of different types of brain tumours, acquired in vivo from 334 patients at four different centres, are clustered according to their pathology, using automated pattern recognition techniques and the results are presented as a two-dimensional scatterplot using an intuitive graphical user interface (GUI). Formal quality control procedures were performed to standardize the performance of the instruments and check each spectrum, and teams of expert neuroradiologists, neurosurgeons, neurologists and neuropathologists clinically validated each case. The prototype decision support system (DSS) successfully classified 89% of the cases in an independent test set of 91 cases of the most frequent tumour types (meningiomas, low-grade gliomas and high-grade malignant tumours-glioblastomas and metastases). It also helps to resolve diagnostic difficulty in borderline cases. When the prototype was tested by radiologists and other clinicians it was favourably received. Results of the preliminary clinical analysis of the added value of using the DSS for brain tumour diagnosis with MRS showed a small but significant improvement over MRI used alone. In the comparison of individual pathologies, PNETs were significantly better diagnosed with the DSS than with MRI alone.

No abstract

Justification Automatic brain tumor classification by MRS has been under development for more than a decade. Nonetheless, to our knowledge, there are no published evaluations of predictive models with unseen cases that are subsequently acquired in different centers. The multicenter eTUMOUR project (2004)(2005)(2006)(2007)(2008)(2009), which builds upon previous expertise from the INTERPRET project (2000INTERPRET project ( -2002 has allowed such an evaluation to take place. Materials and Methods A total of 253 pairwise classifiers for glioblastoma, meningioma, metastasis, and low-grade glial diagnosis were inferred based on 211 SV short TE INTERPRET MR spectra obtained at 1.5 T (PRESS or STEAM, 20-32 ms) and automatically pre-processed. Afterwards, the classifiers were tested with 97 spectra, which were subsequently compiled during eTUMOUR. ResultsIn our results based on subsequently acquired spectra, accuracies of around 90% were achieved for most of the pairwise discrimination problems. The exception was for the glioblastoma versus metastasis discrimination, which was below 78%. A more clear definition of metastases may be obtained by other approaches, such as MRSI + MRI. Conclusions The prediction of the tumor type of in-vivo MRS is possible using classifiers developed from previously acquired data, in different hospitals with different instrumentation under the same acquisition protocols. This methodology may find application for assisting in the diagnosis of new brain tumor cases and for the quality control of multicenter MRS databases.

Automated pattern recognition techniques are needed to help radiologists categorize MRS data of brain tumors according to histological type and grade. A major question is whether a computer program "trained" on spectra from one hospital will be able to classify those from another, particularly if the acquisition protocol is different. A subset of 144 histopathologically validated brain tumor spectra in the INTERPRET database, obtained from three of the collaborating centers, was grouped into meningiomas, low-grade astrocytomas, and "aggressive tumors" (glioblastomas and metastases). Spectra from two centers formed the training set (94 spectra) while the third acted as the test set (50 spectra). Linear discriminant analysis successfully classified 48/50 in the test set; the remaining two were atypical cases. When the training and test sets were combined, 133 of the 144 spectra were correctly classified using the leaveone-out procedure. These spectra had been obtained using different sequences (STEAM and PRESS), different echo times (20, 30, 31, and 32 ms), different repetition times (1600 and 2000 ms), and different manufacturers' instruments (GE and Philips). Pattern recognition algorithms are less sensitive to acquisition parameters than had been expected. Magn Reson