3. MARCO TEÓRICO
3.2 El comedor escolar
3.2.1 Planificación y gestión
Performed in collaboration with Ferran Prados Carrasco (joint first author); plus Arman Eshaghi, Carole Sudre, Tom Button, Matteo Pardini, Rebecca Samson, Sebastien Ourselin, Claudia Gandini Wheeler-Kingshott, Joanne Jones, Alasdair Coles, and Declan Chard.
My role: conception and design of study; gathered existing scans and clinical information (including updating 13-year alemtuzumab data from research clinic); planned and performed image processing pipeline (with Ferran Prados Carrasco); quality-assurance steps; analysed data (with assistance from Arman Eshaghi); wrote manuscript and amended following co-author and journal reviews.
Participants
This study utilised the same early relapsing-remitting multiple sclerosis cohorts as Button et al., 2013. Twenty alemtuzumab-treated patients from the CAMMS223 trial (ClinicalTrials.gov identifier: NCT0050778) were scanned immediately before treatment then annually for 3 years (scans performed between 2003-2007). Twenty-two people from a natural history cohort with annual imaging were scanned three times between 1998-2003 (Davies et al., 2005). In the alemtuzumab cohort, clinical follow-up (including Expanded Disability Status Scale (EDSS) scores (Kurtzke, 1983)) was undertaken every 3 months for 3 years then annually thereafter.
Ethical approval was granted by the joint ethics committee for the Institute of Neurology and the National Hospital for Neurology and Neurosurgery, Queen Square, London (for the natural
history cohort) and by local review boards and ethics committees (for the CAMMS223 cohort).
Written informed consent was obtained from all participants.
Magnetic resonance imaging
All images were acquired using the same 1.5 T Signa scanner (General Electric, Milwaukee, USA) at the Institute of Neurology, University College London using the same protocol (Davies et al., 2005). The acquisition included (i) dual echo T2-weighted/PD images of the whole brain (50x3mm contiguous axial-oblique slices parallel to the anterior/posterior commissural line, matrix 256x256, FOV 220x220mm, TR=3200ms) for lesion-identification and segmentation; (ii) T1-weighted spin-echo whole-brain images (50x3mm contiguous axial-oblique slices parallel to the anterior/posterior commissural line, matrix 256x256, FOV 220x220mm, TR=600ms, echo time=17ms) for segmentation; and (iii) 2D interleaved dual-echo, spin echo magnetisation transfer imaging of the whole brain (28x5mm contiguous axial-oblique slices parallel to the anterior/posterior commissural line with an interleaved sequence (TR=1720ms, echo time=30/80ms, number of excitations 0.75, matrix 256x256, FOV 240x240mm, magnetisation transfer-weighted by the application of pre-saturation pulse (Hamming apodized 3 lobe sinc pulse, duration 64ms, flip angle 1430°, peak amplitude of 14.6mT giving a normal bandwidth of 62.5Hz, applied 1 kHz from the water resonance)).
The scanner was upgraded in April 2004; statistical analyses were adjusted to account for the small resultant MTR step-increase.
Image analysis
WM lesions were outlined on PD/T2-weighted images by TB using JIM v5.0 (Xinapse systems, Aldwincle, UK). The WM lesion mask was then resampled from PD/T2 to T1-weighted image
space using a transformation obtained by registering the pseudo-T1 image (generated from the PD/T2-weighted image (Hickman et al., 2002)) to the weighted image. Lesion filled T1-weighted images (Popescu et al., 2014; Prados et al., 2016) were segmented into WM, cortical grey matter (GM), deep GM and CSF using the GIF algorithm (Cardoso et al., 2015) and tissue masks were binarised with a 90% probabilistic threshold (Brown et al., 2017). These segmentations were also used to calculate brain parenchymal fraction, calculated as follows: (GM volume + WM volume) / (GM volume + WM volume + CSF volume).
Lesions were dilated by 2 voxel layers to remove perilesional MTR abnormalities (Vrenken et al., 2006), then subtracted from the WM segmentations to produce NAWM masks. MTR maps were calculated in T1-weighted image space on a voxel-by-voxel basis as follows: MTR (in pu) = (((MToff
-MTon) / MToff) x100) using the short-echo data because of its higher signal-to-noise ratio.
The whole brain was segmented into 12 concentric bands between the ventricular walls and pial surfaces using the normalised distance map derived from the Laplace equation isolines between the two surfaces (Yezzi and Prince, 2003; Pardini et al., 2016). The NAWM mask was superimposed over these bands (removing the deep GM (DGM) and cortical GM) then applied to the MTR maps, generating 12 bands of NAWM. Consistent with previous work using 2D MTR data (Brown et al., 2017), the first two (periventricular) bands were excluded to mitigate partial-volume effects with CSF (Figure 4).
Figure 4: Brain segmentation into 12 concentric bands using the iterative application of the normalised central curve of the Laplace equation. The first 2 periventricular bands were excluded to mitigate partial volume effects, leaving 10 bands. Each band is represented by a different colour.
Statistics
Mean and standard error MTR (in pu) were calculated in each band, the periventricular NAWM (over the first 3 bands), the whole brain NAWM, the DGM and the cortical GM. Mean and standard error periventricular MTR gradients were calculated over the first three bands adjacent to the ventricles ((MTR in NAWM band 3 – MTR in NAWM band 1)/2) as previously described (Pardini et al., 2016). Higher periventricular gradients are more abnormal; while higher mean MTR values are less abnormal.
Nested mixed-effects models adjusted for age, gender, prior relapse rate, BPF, scanner upgrade status and multiple comparisons compared the change in (i) periventricular NAWM MTR
gradients; and for comparison (ii) mean whole brain NAWM MTR and (iii) mean DGM MTR between the alemtuzumab and untreated groups. Further mixed-effects models adjusted for the same covariates compared the baseline (i) periventricular MTR gradient, (ii) mean whole brain NAWM MTR and (iii) mean DGM MTR between those who did or did not relapse following treatment with alemtuzumab. We used a 4-year cut-off (reflecting the time when half of those that relapsed had done so); and performed sensitivity analyses using different cut-offs. We previously reported the predictive value of the periventricular MTR gradient in untreated patients following a clinically-isolated ON (Brown et al., 2017), in which 3D MTR was used in more patients (n=71) compared to the present study. The present study necessarily excluded patients in the untreated cohort that began disease-modifying treatment before their finals scan.
However, for completeness, we repeated our analyses in this smaller untreated cohort and report them seperately (p.Error! Bookmark not defined.).
To determine whether or not changes in the periventricular MTR gradient were independent of lesional changes, we repeated each statistical model additionally adjusting for change in WM lesion number (first adjusting for change in whole brain lesion number; then adjusting for change in periventricular lesion number (calculated over the first 3 bands)). To explore whether changes in periventricular MTR gradient were distinct from more diffuse MTR changes, we additionally adjusted each MTR gradient model for change in periventricular NAWM MTR and whole brain NAWM MTR; and each mean whole brain NAWM MTR model for change in periventricular MTR gradient.
Spearman rank statistics explored the correlations between (i) changes in whole brain NAWM MTR and periventricular MTR gradient (to further examine the relationship between the two metrics); and (ii) periventricular gradient and EDSS score. Statistics were performed in R (v3.3.1).
Results were considered statistically significant at the p<0.05 level.