Lukas Lundholm1, Mikael Montelius1, Oscar Jalnefjord1, Eva Forssell-Aronsson1, and Maria Ljungberg1
1Department of Radiation Physics, Institute of Clinical Sciences, Gothenburg, Sweden
Synopsis
VERDICT
dMRI allows for estimation of microstructural parameters in tumours which could
facilitate planning and assessment of treatment. Due to the complexity of the
model there is a risk for overfitting to data and there is hence a need to
determine the reliability of the estimated parameters. A mouse model of human SI-NETs
(n=5) was measured twice using the same dMRI protocol and the VERDICT model was
fitted to data. Results showed an overall good repeatability of the tumour mean
parameter values estimated by VERDICT. However, some local clusters of voxels
showed larger differences between repeated scans.
Introduction
Non-invasive
methods to estimate the tissue microstructure in tumours could guide the choice
of treatment and be used to assess treatment effects. Diffusion MRI (dMRI) is a
technique that allows for probing of the microstructure by making the MR signal
sensitive to the motion of water molecules. By fitting mathematical models to
the dMRI data, it is possible to indirectly estimate microstructural properties
of the tissue. The Vascular, Extracellular, and Restricted Diffusion for Cytometry
in Tumours (VERDICT) model is designed to estimate the voxel volume fractions
of intracellular (fIC), vascular (fVASC) and
extracellular extravascular (fEES) space, as well as the cell radius
(R).1 VERDICT has shown promising results
for diagnosing cancer and assessment of treatment.2,3
However, due to the complexity of the model there is a risk of overfitting and
unreliable model parameter estimations. To make the model fitting more robust it
is therefore common to fix some of the model parameters and to use
regularization methods. It is hence important to determine the reliability of
the parameters estimated by VERDICT in different tumour tissues.
The aim of
this study was to investigate the repeatability of VERDICT dMRI in a model of human
small intestine neuroendocrine tumours (SI-NET).Subjects and methods
MRI experiments
were performed on Balb/C nude mice (n=5) with subcutaneous xenografts of human SI-NETs
(GOT1-1) using a 7T preclinical MR system (Bruker, Biospec). Each animal was
imaged twice using the same dMRI sequence, without moving the animal (scan
protocol parameters in Table 1). Anaesthesia was maintained using 2-2.5 % isoflurane
in air (Isoba vet., Schering-Plough Animal-health, Denmark) delivered via a
nose cone. The VERDICT model fitting for estimation of R, fIC, fVASC,
and fEES was performed using the AMICO framework.4 The BallSphere model was used for the
intracellular and extracellular extravascular compartment, and the vascular
compartment was explained using a model which separates blood flow from blood
diffusion.5
The diffusion coefficients (D) of each compartment as well as the velocity
dispersion (vd) of the blood flow were fixed to increase the robustness of the
fit (DIC=1x10-9 m2/s,
DEES=1.5x10-9 m2/s, DVASC=1.75x10-9
m2/s, vdVASC=0.6x10-3 m/s).
To evaluate
the repeatability of the method the repeatability coefficient (RPC) and
intraclass correlation coefficient (ICC) were calculated for each estimated
parameter. The RPC was calculated according to
$$RPC=1.96SD$$
where SD was the
standard deviation of the differences between repeated measurements. The ICC
and its 95% confidence interval were calculated using the ICC(A,1) formula.6
This study
was approved by the Gothenburg Ethical Committee on Animal Research.Results and discussion
An overall
good agreement between repeated measurements was seen for the mean parameter
values in the tumour (Figure 1). The fIC parameter showed the
highest ICC (0.97) with a high lower bound of the 95% confidence interval
(0.80). The fVASC and fEES parameters also showed good
ICCs (0.80 and 0.86) but with poorer lower bounds of the confidence interval
(0.10 and 0.13). All parameters representing volume fractions showed similar
RPCs (0.04 – 0.06) suggesting that the absolute differences between repeated
measurements were similar for these parameters. However, because fIC
was estimated as much higher than fVASC and fEES in the studied
tumours the relative differences between repeated measurements were lower for
the fIC parameter, as reflected by its higher ICC. The results also
showed that estimation of the R parameter may be less reliable (ICC=0.24) with
expected differences of about 0.14±0.53 micrometres between repeated measurements
in the studied tumour tissue.
Some larger
differences were seen when comparing the repeated measurements voxel wise (Figure
2). Although most colormap representations of the estimated parameters appeared
similar on repeated experiments, some regional clusters of voxels within the
tumours showed substantially changed values. Such regional changes within the
tumour was seen for all parameters but was the most prevalent for the R
parameter. There may have been some biophysical changes within the tumour
tissue during the measurements which could have affected the dMRI signal and
thus the parameter estimations. However, such changes are unlikely to fully
explain the substantial differences observed in some regions of the VERDICT
parameter colormaps.Conclusion
Our results
showed that VERDICT dMRI has an overall good repeatability in the studied tumour
model for the tumour mean parameter values. However, VERDICT showed a large
regional sensitivity to signal changes between scans which led to poorer voxel
wise test-retest reliability of the estimated parameters. Further studies are
required to test the reliability of the parameters estimated by VERDICT for
different tumour tissue types, scan protocols, and model fitting methods.Acknowledgements
No acknowledgement found.References
- Panagiotaki et al., NRM in Biomed. 2014
- Johnston et al., Radiology. 2019
- Roberts et al., Sci Rep. 2020
- Bonet-Carne et al., NMR in Biomed. 2018
- Ahlgren et al., NMR in Biomed. 2016
- McGraw et al., Psychological Methods. 1996