Speaker
Description
Summary
The SPM software library was originally developed and is made freely available by
the Functional Imaging Lab (FIL) at the Wellcome Department of Imaging Neuroscience
(London University College) for activation studies in functional MR.
Since then, the use of SPM was extended and, through a specifically defined
analysis protocol, SPM routines are presently the standard within the neurological
research community as regards a voxel based analysis of PET/SPECT studies for the
early diagnosis of AD.
In order to achieve correct results, the SPM software library provides a number of
functionalities related to image processing and statistical analysis:
normalization, co-registration, smoothing, parameter estimation, statistical
mapping.
The statistical parametric mapping algorithm (the most important functionality for
our goal) performs a statistical analysis in order to compare, on a voxel-by-voxel
base, the perfusion values in the test images against the corresponding values in
normal images.
As a result of a previous research project [1], remote access to SPM is being made
available through the Italian Portal of Neuroinformatics providing doctors from
peripheral hospitals with an invaluable tool to increase the “comparison database”
and therefore improving the AD diagnosis. The portal contains a section entirely
dedicated to the statistical analysis of PET/SPECT images, accessible by authorized
users only. Doctors or researchers accessing the portal may thus be supported in
running analysis tasks on suspect AD patient studies. Directly from the portal, a
user can upload the suspect AD image and select the normal cases for statistical
calculation.
The SPM application is available to authorized users without downloading any
software tool. In order to use it, no particular hardware resource or specific
computer knowledge is needed.
In order to evaluate the potential advantage of porting such an implementation to a
Grid environment, it is worth noting that during the statistical parametric mapping
a large set of images of normal patients is required to be used for comparison.
This is because the accuracy of ipoperfusion maps is strictly related to the number
of normal studies compared to the test image.
On the other hand, due to ethical issues and to the high costs of neuroimaging
technologies, PET and SPECT studies on normal subjects are very rare. The NEST-DD
project, funded by the European Commission [2], collected a database of about 100
images in order to make available the first large dataset for these studies.
Moreover the images of normal subjects are covered by privacy and security issues
and for this reason they cannot be freely moved on the net or published by the
centre that made the analysis. As a consequence, only doctors working at very large
institutions, locally owning large databases of normal images, can usually carry
out SPM-based analyses.
Starting from these considerations, the aim of our project has been to enable
doctors from small peripheral hospitals to use large sets of normal PET/SPECT
images provided by medical research institutes distributed on the net, by remotely
extracting the information needed for the statistical analysis from the normal
images and collecting it without moving the original image files.
Furthermore, the execution time of the analysis must be compatible with an
interactive clinical application in a busy medical environment. The time required
for the analysis can be reduced, since:
- some aspects of calculation could be parallelized and distributed on the
computational resources associated to the remote databases of normal images;
- the time required for data transfers over the network would be reduced, since the
code amounts to just few KB, compared to images sizing up to 100 MB.
The use of GRID technologies well matches all of the above issues and allows easy
access to distributed data as well as to distributed computational resources.
Grid implementation has been carried on GILDA Infrastructure that provides a series
of sites and services spread all over Italy and the rest of the world on which LCG
and gLite middleware are installed and several Virtual Organisations are enabled.
A LCG node has been installed at University of Genoa for the implementation of
biomedical applications and in particular for this application.
The objectives of the LCG-based implementation are:
- to distribute PET/SPECT images on different storage resources available on the
GRID and register them on a catalogue.
- to insert and manage metadata in order to make the user able to select normal
images for the statistical analysis using their own attributes.
- to access images from User Interface using Logical File Names (LFN) without
moving them from storage resources.
To reach the first result, LCG File Catalog (LFC) was selected: it allows users and
applications to locate files (or replicas) on the LCG Grid maintaining mappings
between logical and physical file names.
As next step, the ARDA Metadata Grid Application (AMGA) has been integrated,
fulfilling also the second requirement. Actually, LCG does not provide a
satisfactory metadata management system and AMGA fills this hole. The collected
metadata are associated to files stored on the LCG Grid through a reference on the
LFC catalogue system and are used to select images directly through the portal.
AMGA provides the ability to allow only certain people to access specified
attributes. This is very important because all medical data should be considered as
sensitive to preserve patient privacy.
To meet the third requirement, LCG Data Management and File access tools have been
selected.
In particular lcg_util and Grid File Access Library (GFAL) tools was used to
transparently interact with LFC catalog and to perform calls for storage management
and file access using the correct protocol for file transfer.
[1] S. Scaglione, I. Castiglioni, E. Molinari, F. Cesari, F. Repetto, A. Schenone,
J.Abutalebi, D. Perani, M.C. Gilardi and F. Beltrame, “Neuroinformatics portal as
knowledge repository and e-service for neuroapplication and data mining”, Medicon
2004 (Mediterranean Conference on Medical and Biological Engineering), Naples, July
31-August 5 2004
[2] K. Herholz, E. Salmon, D. Perani, J-C. Baron, V. Holthoff, L. Frolich, P.
Schonknecht, K. Ito, R. Mielke, E. Kalbe, G. Zundorf, X. Delbeuck, O. Pelati, D.
Anchisi, F. Fazio, N. Kerrouche, B. Desgranges, F. Eustache, B. Beuthien-Baumann,
C. Menzel, J. Schroder, T. Kato, Y. Arahata, M. Henze, and W-D.
Heiss “Discrimination between Alzheimer Dementia and Controls by Automated Analysis
of Multicenter FDG PET” NeuroImage 17, 302–316 (2002)
