Physica Medica: European Journal of Medical Physics - Volume 32 - Supplement 1 - February 2016

131 / 146

E.425
ANALYSIS OF SIMULTANEOUS EEG/FMRI DATA ACQUISITION IN EPILEPTIC
PATIENTS: COMPARISON OF SEMI-BLIND ICA, SPATIAL ICA AND GLM
BASED METHODS
M. Carnì C. Di Bonaventura C. Borrazzo J. Fattouc A.T. Giallonard
S. Casciato A. Morano E. Di Castr C. Colonnes a UOC Medical Physics,
Policlinico Umberto I, Rome, Italy; b Department of Neurology and Psychiatry,
University of Rome Sapienza, Rome, Italy; c Department of Molecular Medicine,
University of Rome La Sapienza, Rome, Italy
Introduction: A simultaneous recording of electroencephalogram (EEG) and
functional magnetic resonance imaging (fMRI) is a powerful and promis-
ing tool in epilepsy. In this work, we compared a two data driven analysis
based on spatial independent component analysis (ICA) and semi-blind ICA
(constrained analysis) with a traditional approach, general linear model
(GLM). We applied these approaches on simultaneous EEG/fMRI data ac-
quired on patients affected by ictal electro-clinical activity.
Materials and Methods: FMRI was performed on a clinical 1.5 T magnet
(Philips Gyroscan Intera). Two GE-EPI scans, each made up of 200 tempo-
ral dynamics, with each dynamic consisting of 20 axial slices (5 mm
thickness, matrix 64 × 64, FOV 24 × 24 cm2, TR 3000 ms, TE 50 ms). EEG re-
cording was performed using a 32-channels MR-compatible device
(Micromed, Italy) and cleaned of MRI gradient and ballistocardiographic
artefacts. Ten patients with epilepsy participated and gave informed consent.
The fMRI data were analysed using three methods; information extracted
from EEG data was used. Spatial ICA and Semi-Blind ICA was performed
with Group ICA MATLAB (MathWorks,Inc) Toolbox
. GLMwas performed with SPM8 . A cross-
correlation analysis was performed between the relative time-course of
Independent Components (ICs) map and a GLM regressor.
Results: ICA and semi-blind ICA results show comparable to activation areas
obtained by GLM analysis, in agreement with presumed electroclinical hy-
pothesis. The semi-blind ICA results show higher correlation value with GLM
regressor.
Conclusion: We have demonstrated that the same BOLD patterns of acti-
vation in response to synchronized ictal activity were found by ICA, Semi-
blind ICA and GLM analysis. Semi-blind ICA improved the power of ICA in
the presence of noise. The selection of these ICs interesting components
remains the main problem of the application of data-driven approach to
EEG/fMRI data.
E.426
HIGH-FIELD MR SPECTROSCOPY IN THE MULTIPARAMETRIC MRI
EVALUATION OF BREAST LESIONS
C. Cavedon * G. Meliado’ L. Camera I. Baglio F. Caumo
S. Montemezz a Azienda Ospedaliera Universitaria Integrata, Verona, Italy;
b ULSS 20 Verona, Verona, Italy
Introduction: To assess the role of 3T-MR spectroscopy (MRS) in the multi-
parametric MRI evaluation of breast lesions, using a pattern-recognition
based classification method.
Materials and Methods: 291 patients (301 lesions, range 0.34–115.45 cm3,
mean 8.2 cm3, median 2.3 cm3) were enrolled in the study (February 2012–
July 2015, age 18–85 y, mean 54.2 y). T1-TSE (TR/TE = 400/10 ms) and T2-
STIR imaging (TR/TE = 5000/60 ms), dynamic-contrast-enhanced MRI
(DCE-MRI), apparent diffusion coefficient (ADC) (b = 0–800 sec/mm2), and
single-voxel MRS (10 × 10 × 10 mm3, pencil-beam shimming, PRESS,
TR/TE = 3000 ms/135 ms) were performed by means of a 3T scanner (Philips
Achieva STx). A RF multiple-source systemwas used to mitigate artifacts due
to the relatively short RF wavelength at 3T. MRS results were accepted if the
FWHM of the water peak was <45 Hz. Total choline (tCho) was considered
detected if the signal-to-noise ratio (SNR) of the 3.2 ppm peak was >2. A
classifier-based analysis (support-vector-machines, SVM) was performed with
4-dimensional vectors including type of margin, DCE-MRI kinetic curve class,
ADC mean value, and tCho SNR. A comparison with 3-dimensional vectors
(excluding MRS) was performed to assess MRS impact on sensitivity, speci-
ficity, and positive-negative predictive values (PPV-NPV) for malignancy.
Results: 228 lesions (180 malignant / 48 benign) showed acceptable spec-
tral quality. Comparison of classification results with histopathological
examination of surgical specimens (or micro-biopsy for 29/48 benign lesions)
showed sensitivity = 93.7% (95% C.I. = 88.3–97.2%), specificity = 84.9% (71.2–
93.6%), PPV = 95.2% (90.2–98.0%), NPV = 81.5% (67.6–91.1%) without the
inclusion of MRS in SVM analysis. When MRS was included, the figures in-
creased to 95.1% (C.I. 90.3–98.1%), 90.7% (78.0–97.2%), 97.2% (93.0–99.1%),
and 85.0% (71.3–93.6%), respectively.
Conclusion: Inclusion of 3T-MRS in the multi-parametric MRI evaluation
of breast lesions improved the performance of the SVM-based classifier.
E.427
FIBROSING LIVER DISEASES IN PAEDIATRIC AGE: MRI INVESTIGATION BY
DIAGNOSTIC MAPS OF SLOW DIFFUSION AND FAST DIFFUSION
GENERATED FROM A MULTIPLE B VALUES DWI
A. Ciccarone P. Gulin M. Esposit C. Defilippi a Azienda Ospedaliero-
Universitaria Meyer, Firenze, Italy; b Azienda Sanitaria ASF10, Firenze, Italy
Introduction: Our work focuses on generate and evaluate diagnostic maps
of slow diffusion and fast diffusion (likely perfusion) through IVIM model
in children, comparing the results in patients with those in healthy subjects.
Methods and Materials: We employed the bi-exponential algorithm of Le
Bihan to study the two compartments of fast diffusion (f) and slow diffu-
sion (D). Home-made MATLAB was developed to generate maps of IVIM
parameters. We studied two groups of paediatric patients, one of 5 healthy
volunteers and one of 10 patients with liver disease with possible fibrotic
evolution. All patients underwent liver MRI (with acquisition of a respiratory-
triggered DWI sequence with 11/13 b values) and biopsy. For each patient,
four maps were extracted: ADCf (D, slow diffusion coefficient), PPC (D*,
pseudodiffusion coefficient), PFC (f, perfusion fraction) and Flow (fxD*). ROIs
have been placed on each map in each liver segment; their values were
extracted and included in the statistical analysis.
Results: P-value <0.05 was considered statistically significant. The mean
values (with SD) in the group of patients are: D = 0.814 ± 0.08 × 10-
3 mm2/s (P < 0.001), D* = 93.69 ± 37 × 10-3 mm2/s (P = 0.024), f = 16.9 ± 4.1
% (P = 0.041), Flow = 163.4 ± 86.5 × 10-3 mm2/s (P = 0.012).
Conclusions: All results are statistically significant. In particular, the D pa-
rameter shows the highest reproducibility and the lower standard deviation
in our population, suggesting a promising use of slow diffusion coeffi-
cient to evaluate the diffusion in the liver and the consequent quantification
of fibrosis.
E.428
ACCURACY OF THE GRASE SEQUENCE IN EVALUATING T2 RELAXATION
TIME
A. Coniglio D. Landi S. Vollaro D. Lupoi E. Belligott L. Begnozz
a Medical Physics Unit, S. Giovanni Calibita Fatebenefratelli Hospital, Roma, Italy;
b Medicina dei Sistemi Department, Medicine and Surgery Faculty, ‘Tor Vergata’
University, Roma, Italy; c Department of Neurology, Campus Bio-Medico
University of Rome, Roma, Italy; d Department of Radiology, Fatebenefratelli
Hospital, S. Giovanni Calibita, Roma, Italy; e Specialization school of Medical
Physics, Tor Vergata University, Roma, Italy
Introduction: Over the last few years, GRASE (gradient and spin echo) se-
quence has been used to assess the iron content in multiple sclerosis patients.
The present study aims to assess the accuracy of the GRASE sequence in
the evaluation of T2 relaxation times.
Materials and Methods: Acquisitions were performed with a dedicated
phantom composed of seven vials containing manganese solutions with
salt concentration in the range 0.10–1.20 mmol. Multiple spin echo se-
quence with echo time ranging from 50 to 400 ms were used to measure
T2 relaxation times of the solutions (T2 reference). Phantom images were
acquired using GRASE and T2 values were calculated with an in-house code,
using amono-exponential fitting function (T2GRASE). The obtained T2 values
were compared with those automatically calculated from the proprietary
magnetic resonance software (T2 Software). Acquisitions with GRASE were
also performed on 10 controls. Mean T2 values were calculated for thala-
mus, putamen, pallidus, caudatus and hippocampus. Bland–Altman plot were
calculated to analyze the agreement between T2 GRASE and T2 Software.
e126 Abstracts/Physica Medica 32 (2016) e124–e134