This poster is published under an
open license. Please read the
disclaimer for further details.
Keywords:
Abdomen, Colon, Gastrointestinal tract, MR-Diffusion/Perfusion, Technical aspects, Cancer
Authors:
G. C. Manikis1, K. Marias1, K. Nikiforaki1, D. M. J. Lambregts2, M. V. Heeswijk2, R. G. H. Beets-Tan2, N. Papanikolaou3; 1Heraklion/GR, 2Maastricht/NL, 3Stockholm/SE
DOI:
10.1594/ecr2016/C-2178
Methods and materials
Sixteen consecutive patients with rectal adenocarcinoma underwent MRI examination before chemoradiation therapy or surgery.
Diffusion weighted imaging (DWI) using a Spin-Echo Echo Planar Imaging sequence was acquired on a Philips Ingenia 1.5T system utilizing 6 b-values (0,
25,
50,
100,
500,
1000) assymetrically sampled to acquire more data on the low b-value area were microperfusion effects are more evident.
Mono-exponential and bi-exponential diffusion models were used both in Gaussian and non-Gaussian form to fit the data and extract the relevant diffusion imaging biomarkers.
ADC was calculated based on the mono-exponential Gaussian model,
while two components were used in the bi-exponential Gaussian model,
the true Diffusion and Pseudo-Diffusion,
leading to the quantification of f (microperfusion fraction),
D* (pseudo diffusion coefficient) and D (true diffusion coefficient).
D(K) (non-Gaussian Diffusion coefficient),
and K (kurtosis) were calculated based on the mono-exponential non-Gaussian model,
while the bi-exponential non-Gaussian model was used to extract f(k),
D*(K),
D(K) and K.
All models were applied on multiple regions of interest including rectal tumors and their associated biomarkers were derived for every pixel in the region using non-linear fitting techniques.
Four different statistical criteria were recruited to determine which diffusion model provided more accurate fitting on the experimental data.
These statistical criteria were adjusted R-square (adj-R2) ,
root mean square (RMSE),
sum of squares due to error (SSE) and corrected Akaike information criterion (cAIC) [4].
Wilkoxon signed-rank and Dunn's non-parametric statistical tests were used to disclose any significant differences between all four models.