IST tumour

Document Sample
IST  tumour Powered By Docstoc




We aim to facilitate the use of magnetic resonance spectroscopy (MRS) for improved
diagnosis and therapy of patients with brain tumours and other space occupying lesions.
Spectra, easily acquired alongside commonplace MR imaging (MRI) procedures, uniquely
delineate biochemistry of human tissue in situ. Although MRS gives significantly
improved brain tumour categorisation, it is not widely used, partly because radiologists
have difficulty in interpreting spectral data. We therefore aim to develop a user-friendly
computer program for spectral classification. Systems development will be informed by
(a) a large "training set" of data contributed by members of the consortium and (b) new
spectra acquired under agreed protocols. Automated pattern recognition techniques will be
developed for tumour classification together with an intuitive interface. The program will
promote wider use of MRS, hence reducing the need for distressing and dangerous brain

Advisory Group
Brain tumours
Decision Support Tool
Graphical user interface (GUI)
Magnetic resonance imaging (MRI)
Magnetic resonance spectroscopy (MRS)
Nosologic image
Pattern recognition (PR)
The goal of this project is to develop a computer-based decision support tool, installed in
hospital MRI centres, that will enable radiologists and other clinicians without special
knowledge or expertise to diagnose and grade brain tumours routinely using magnetic
resonance spectroscopy.
MRS is coming to have a role in routine clinical medicine, but a major drawback is
difficulty of interpretation of the spectra. All suitable MRI/MRS instruments are situated
in radiology departments, and used for mainly anatomical MRI studies. MR spectra, with
their rich content of metabolic information, require a totally different kind of
interpretation; the aim of INTERPRET is to develop and test an automated, objective
system that will allow radiologists to diagnose and grade brain tumours entirely on the
basis of MR spectra.

This project will contribute to the strategic objectives of the IST programme, which is
aimed at realising the benefits of the information society for Europe. INTERPRET will
develop a program that will enable physicians treating patients with brain tumours to
obtain quicker, more accurate and entirely objective diagnoses without the need for brain
biopsy. It will use the highly advanced MRS technology that has been developed as an
ancillary to MR imaging in Europe in the last 20 years. This knowledge-based system will
enhance the applicability of the MRS technique and accelerate its take-up in Europe.
    A suspected brain tumour presents a difficult challenge to the physician and a devastating
    finding for the patient. The type and grade (degree of malignancy) of the cancer must be
    confirmed before a treatment plan can be developed and this almost always requires
    histopathological examination of a biopsy. Brain biopsy is invasive, has significant
    morbidity and mortality, and does not always give the right answer. One problem is that
    only a small fragment of an often heterogeneous tumour can be taken. Furthermore,
    histopathological evaluation (even of a large surgical specimen) is inevitably subjective,
    demanding skill and experience. There has been thus much recent interest in the
    automatic, objective diagnosis and grading of brain tumours using information from 1H
 H MRS is a non-invasive technique for measuring the chemical content of living tissues. Data
are displayed as spectra, and the peaks give a measure of the different chemicals within the
tissue. 1H MRS can be performed on most 1.5 Tesla clinical MR instruments, many of which
have highly developed programs for obtaining spectra from single volumes (voxels) or arrays of
(multi) voxels (termed Spectroscopic Images or Chemical Shift Images) in the brain. 1H MRS
has been widely applied for studying brain tumours, which show marked differences in their
spectral patterns that correlate with the tumour type and grade (see Figure 1).

    A spectrum (or set of multi-voxel spectra) can be obtained in approximately ten minutes
    and many existing hospital MR instruments can acquire MR images and spectra during the
    same examination. However, few centres presently use MRS, since radiologists find
    spectra difficult to interpret. There are two main reasons for this:
     The chemical information in a spectrum is entirely different from the anatomical
       information provided by most imaging modalities, such as MRI, and few radiologists
       have the necessary biochemical expertise to interpret MRS.
     Differences in spectral patterns between certain classes of tumour are quite subtle and
       there are many borderline cases in all classes.
    For these reasons MR spectra require computer-based methods for accurate detection.

     Month 6: An organised collection of spectra and associated clinical data, suitable for
      use in the first prototype classifier system
     Month 18: A fully operational database management system to store the spectra and
      associated clinical information
     Month 24: A Prototype Classifier System.
     Month 36: The Decision Support Tool, and the Database

    This project will:
    1. Enable radiologists to categorise brain tumours using MRS.
    2. Aid planning of treatment and therapy.
    3. Alleviate patient distress.
    4. Facilitate the uptake of MRS by clinicians.
    5. Consolidate MRS as a viable alternative to brain biopsy.
There has been much recent interest in the automatic, objective diagnosis and grading of
brain tumours based on non-invasive MR Spectroscopy. So far, there has been some
success from programs based on small training datasets obtained at single institutions,
but several of the published ones require significant input from the operator in the form of
peak picking, which weakens their objectivity and also makes them much less practical in
routine operation. INTERPRET aims to go beyond these programs in the following
1. The Database
We will collect a much larger dataset of spectra from tumours than any currently available
(from approximately 900 patients), and will ensure that all the spectra admitted have
unequivocal diagnoses, based on the consensus views of the collaborating pathologists and
referring clinicians. The database will be designed (subject to password protection) to
permit any of the collaborating centres to transmit new data by the Internet in a
standardised format, and for any of them to download and read any of the existing data.
This database will be used to make the training dataset required for the final program.

2. Pattern Recognition Methods
 Several studies have been published recently showing good results in applying automated
pattern recognition techniques to MRS data of brain tumours. However, the small number
of spectra available in each study limited the methods that could be used. For the same
reason, it has not been possible to test these methods adequately, or give any measure of
confidence to the results. Because INTERPRET will accumulate enough data, we will be
able to use more advanced methods, and to thoroughly test our classifiers in order that
they can actually be used in a clinical setting.
The classifiers will be entirely automated and objective, requiring no input from the
operator after the initial selection of the Volume of Interest.
3. Evaluation of the Requirement for Standardised MRS Methods
Another innovation will be attention to the problem of assessing spectra obtained under
different conditions. It is not clear how far a database of spectra obtained using one
protocol (or even one particular instrument) can be used to evaluate spectra obtained
under different conditions. A study [1], performed by members of the consortium
using spectra obtained with two instruments using different recycle times, gave promising
results, but more systematic evaluation is required. We will investigate whether studies on
instruments from different manufacturers give compatible results (Siemens, Phillips and
GE instruments will be used in this study), which parts of the protocol can be changed
and which are critical. This innovation will be essential for a practical clinical program as
one cannot expect MRS technology to remain frozen, and it would be impossible to
assemble a new database every time a new pulse program or updated instrument appeared.
Whereas other studies have been restricted to a single protocol, we shall collect data using
four different protocols, considered representative of the protocols generally used in
different centres (two for single voxel spectra and two for multi-voxel spectra) in order
that the performance of the protocols can be compared for different applications.
4. Nosologic “Images”
Until recently, pattern recognition studies were only performed on single-voxel spectra.
Members of our consortium have recently developed a method in which each voxel in
a spectroscopic image is classified and an image is plotted in which regions assigned to
different tumour types, grade, and tissue type are delineated [4]. In an important
innovation, INTERPRET will incorporate a facility for nosologic imaging into
the program, along with a single-voxel capability.

             Figure 2. Nosological image of a high grade glioma with necrosis

5. Graphical User Interface
Another innovation, will be a purpose-designed Graphical User Interface, that will enable
the clinician using the program to (i) input the data required for the evaluation of a new
spectrum, (ii) perform the pattern recognition procedure (which will be automated and
will require no skill or special knowledge), (iii) display the results in a graphical form, and
(iv) see where the new result lies in the data space corresponding to the possible diagnoses
and grades. Nothing at all comparable to this system exists at present.
6. Advisory Group
While INTERPRET is under development we will recruit an Advisory Group consisting of
potential users of the system. Representatives of the instrument manufacturers will also be
invited to join. The initial function of the Advisory Group will be to advise us on the
requirements for the program (particularly the GUI); when a prototype is available,
they will be encouraged to perform beta testing. Once the program is complete, the
Advisory Group will be offered it first, ideally as part of a formal trial.

Hospitals and radiology centres throughout the world. In the short term we will target
centres investigating patients with brain tumours. In the longer term we hope to extend our
system for other types of tumours and diseases.
Participant full name                          Participant acronym Country
Universitat Autònoma de Barcelona              UAB                 Spain
St George’s Hospital Medical School            SGHMS                  U.K
Institut de Diagnòstic per la Imatge           IDI                    Spain
Centre Diagnòstic Pedralbes                    CDP                    Spain
Université Joseph Fourier Grenoble 1           UJF                    France
University of Nijmegen, Faculty of Medical     FMW/AZN                The Netherlands
Katholieke Universiteit Nijmegen               KUN                    The Netherlands
University of Sussex                           UOS                    U.K
PRAXIM, SARL                                   PRAXIM                 France
Siemens AG, Medizinische Technik               SIEMENS                Germany

Dr Carles Arús, Departament de Bioquímica i Biologia Molecular, Universitat Autònoma
de Barcelona, 08193 Cerdanyola del Vallès, Spain

Web page at
If you would like information on joining the Advisory Group please contact
Professor J.R. Griffiths, or Dr. Rosemary Tate,

Related Publications
1. AR.Tate, SJ. Barton, F Howe, JR Griffiths, A. Moreno, I Barba and C. Arús :
Automated classification of in vivo short-echo time 1H MRS spectra of brain tumours
from two different centres using principal component analysis. ISMRM 7th meeting, 1999.
2. AR Tate, JR Griffiths, et al: Towards a method for automated classification of 1H MRS
spectra from brain tumours. NMR in Biomedicine 11:177-191 1998.
3. SJ Barton, FA Howe, SA Cudlip, AR Tate, BA Bell & JR Griffiths: Classification of
human brain tumors with quantitative short echo 1H MRS. Proc. Int. Soc. Magnetic
Resonance in Medicine. 7:426, 1999.
4. F Szabo de Edelenyi, C Rubin, F Estève, S Grand, M Décorps, V Lefournier, J-F Le
Bas, C Rémy: A new approach for analysing proton magnetic resonance spectroscopic
images (1H MRSI) of brain tumors: nosologic images. Nature in Medicine (in press).