Andrada Ianuș1,2, Inês Santiago3,4, Antonio Galzerano3, Paula Montesinos5, Nuno Loução5, Javier Sanchez-Gonzalez5, Daniel C. Alexander2, Celso Matos1,3, and Noam Shemesh1
1Champalimaud Research, Champalimaud Centre for the Unknown, Lisbon, Portugal, 2University College London, London, United Kingdom, 3Champalimaud Clinical Centre, Champalimaud Centre for the Unknown, Lisbon, Portugal, 4Nova Medical School, Lisbon, Portugal, 5Philips Healthcare Iberia, Madrid, Spain
Synopsis
In this study we employed a clinically feasible diffusion MRI
acquisition and modelling approach at 1.5T to characterize perfusion (IVIM) and
higher order diffusion (Kurtosis) properties of mesorectal lymph nodes in
rectal cancer patients upon staging. The results showed that diffusivity
estimated from the IVIM-Kurtosis model was the only parameter showing
significant differences between benign and malignant lymph nodes. Moreover, ROC
analysis evidenced improved differentiation when adding IVIM-Kurtosis to
standard T2-weighted qualitative assessment by expert radiologists.
Introduction
Lymph Node (LN) staging is one of the main determinants of
the management of rectal cancer patients, yet current imaging methods show
limited accuracy for that purpose1. Diffusion MRI (dMRI) is becoming
an increasingly important tool for non-invasive detection of malignant lymph
nodes2-4, with IVIM5,6 and Kurtosis7 models
showing improved results compared to the standard ADC. Combining IVIM and Kurtosis
could further improve LN differentiation8, however, to our
knowledge, it has not been employed so far in mesorectal LN.
This work employs a multi b-value clinical dMRI acquisition
at 1.5T and investigates the ability of ADC, IVIM, Kurtosis and IVIM-Kurtosis
approaches to differentiate between benign and malignant lymph nodes in patients
with rectal cancer. We further compare the performance of dMRI models to
standard T2-weighted qualitative assessment by expert radiologists.Methods
Institutional setting, approved by ethics committee: A total
of 10 patients previously diagnosed with rectal cancer (mean age of 64.9 years,
5 males) were enrolled, after obtaining written informed consent. The imaging
was performed on a 1.5T Philips scanner with the diffusion acquisition added to
the clinical staging pelvic MRI.
Image acquisition: The diffusion data was acquired with
multi-shot Spin-echo Echo-planar Imaging (SE-EPI), with the acquisition
parameters detailed in Table 1. The diffusion acquisition was split in three
parts, with small b-values (50, 100, 200 s/mm2) medium b-values
(500, 1000 s/mm2) and high b-values (1500, 2000, 2500 s/mm2),
to minimize the echo time of the acquisition (TE = 65 ms). In total, 76 lymph
nodes were delineated by a radiologist (8 years experience) and matched to
pathology during macroscopy. Due to motion artifacts, 4 nodes were excluded,
leaving 72 nodes (14 malignant).
Data analysis: Whole-node ROIs were defined on the high
b-value images and copied to the low and medium b-value data sets (Figure 1),
and slightly translated to account for motion when necessary. The average ROI
signal for each b-value was normalized to the ROI average of the b0 images in
the respective set. The normalized signal decay was fitted with four diffusion
models: ADC was fitted to the measurements with medium b-values, IVIM was
fitted to the measurements with small and medium b-values, Kurtosis to the sets
with medium and high b-values and IVIM-Kurtosis to the entire data set.
To investigate the parameter differences between benign and
malignant nodes, we employed a mixed effect linear model with the diffusion
metrics as response variable, and the malignancy status as predictor variable, with
LNs grouped by patient. To assess the differentiation ability of different
models, we employed a ROC analysis.Results
The normalized signal
decay in Figure 1c clearly revealed a faster decay for malignant nodes and a
slower decay for benign nodes, which are also reflected in the model parameters
presented in Figure 2a and Table 2.
Figure 2 shows
box-plots of the parameter values for ADC, IVIM, Kurtosis and IVIM-Kurtosis
estimated in benign and malignant nodes. ADC values were higher in malignant
nodes compared with benign nodes; however, the difference was not statistically
significant (P=0.26). The only parameter which reached a statistically
significant threshold was the diffusivity (D) estimated from the IVIM-Kurtosis
model (P=0.032), which has higher values in the malignant nodes.
Figure 3b presents
the ROC analysis which compared the four diffusion models in terms of
differentiating between benign and malignant lymph nodes. ADC is the worst
performing model, with an area-under-the-curve (AUC) of 0.60, while
IVIM-Kurtosis is the best performing one with an AUC of 0.74, however, the
difference does not reach statistical significance according to the DeLong test
(P=0.16). Figure 3c shows that IVIM-Kurtosis also provides a better
differentiation between benign and malignant nodes (AUC=0.78) compared to the
standard clinical T2 classification based on the ESGAR 2016 criteria (57)
(AUC=0.74 for Reader 1 and AUC=0.59 for Reader 2), which takes into
consideration lymph node size, shape, contour and heterogeneity. Moreover,
combining IVIM-Kurtosis and T2 results significantly improves the
classification compared to T2 weighted images alone (P=0.08 for Reader 1 and
P=0.007 for Reader 2). Discussion
The current study
shows that accounting for IVIM and Kurtosis effects provides better
differentiation between benign and malignant mesorectal lymph nodes compared to
standard ADC and has the potential to improve LN classification when considered
alongside T2 weighted images. One limitation of this study is the relatively
small number of malignant lymph nodes (14/72) and the fact that half of them originate
from only one patient. Nevertheless, this effect was accounted for in the
statistical model. In this study we focused on a ROI analysis to mitigate the effects
of motion, nevertheless in future studies we will also attempt to perform image
registration to enable voxelwise parameter estimation. Conclusion
Quantitative
IVIM-Kurtosis modelling of dMRI data shows potential to improve the
differentiation of benign and malignant mesorectal lymph nodes. Acknowledgements
This study was funded by the
Champalimaud Centre for the Unknown. Dr. Andrada Ianus´ and Prof. Daniel C.
Alexander’s work was also supported by EPSRC grants EP/M020533/1 and
EP/N018702/1 and the NIHR UCLH Biomedical Research Centre. The contributions of Paula Montesino, Nuno
Loução and Javier Gonzales-Sanchez were funded by Philips Healthcare Iberia.References
[1] Grone, J. et al, Accuracy of Various Lymph Node Staging
Criteria in Rectal Cancer with Magnetic Resonance Imaging, J Gastrointest Surg
(2018) 22: 146.
[2] Ogawa, M., et al., The Usefulness of Diffusion MRI in
Detection of Lymph Node Metastases of Colorectal Cancer. Anticancer Research,
2016. 36(2): p. 815-9.
[3] Yasui, O., M. Sato, and A. Kamada, Diffusion-weighted
imaging in the detection of lymph node metastasis in colorectal cancer. Tohoku
J Exp Med, 2009. 218(3): p. 177-83.
[4] Heijnen, L.A., et al., Diffusion-weighted MR imaging in
primary rectal cancer staging demonstrates but does not characterise lymph
nodes. Eur Radiol, 2013. 23(12): p. 3354-60.
[5] Qiu L, et al. Role of quantitative intravoxel incoherent
motion parameters in the preoperative diagnosis of nodal metastasis in patients
with rectal carcinoma. J. Magn. Reson. Imaging. 2016;44: 1031-9.
[6] Yu XP, et al. Discrimination between Metastatic and
Nonmetastatic Mesorectal Lymph Nodes in Rectal Cancer Using Intravoxel
Incoherent Motion Diffusion-weighted Magnetic Resonance Imaging. Acad Radiol.
2016;23:479–85.
[7] Yu, J., et al. Diffusion kurtosis imaging in identifying
the malignancy of lymph nodes during the primary staging of rectal cancer.
Colorectal Dis. 2018;20(2):116-25.
[8] Lu, Y. et al. Extension of the intravoxel incoherent
motion model to non‐gaussian diffusion in head and neck cancer, J. Magn. Reson.
Imaging 2012;36:1088–1096.