The feasibility of performing intravoxel incoherent motion MRI for esophageal cancer and an initial comparison with dynamic contrast-enhanced MRI
Sophie E. Heethuis1, Lucas Goense1,2, Peter S.N. van Rossum1,2, Irene M. Lips1, Richard van Hillegersberg2, Jelle P. Ruurda2, Marco van Vulpen1, Gert J. Meijer1, Jan J.W. Lagendijk1, and Astrid L.H.M.W. van Lier1

1Department of Radiotherapy, UMC Utrecht, Utrecht, Netherlands, 2Department of Surgery, UMC Utrecht, Utrecht, Netherlands


The aim of this study was to investigate whether intravoxel incoherent motion (IVIM) MRI could be a non-invasive alternative for dynamic contrast-enhanced (DCE) MRI for response prediction in patients with esophageal cancer. The feasibility of IVIM was researched, followed by a first comparison with DCE-MRI. It was found that non-rigid registration improved the fitting of the IVIM parameters significantly. Comparison of IVIM and DCE-MRI suggested an inverse relation between perfusion fraction and area-under-the-concentration curve (AUC).


To assess the feasibility and compare IVIM-MRI with DCE-MRI for treatment response assessment in patients with esophageal cancer during neoadjuvant chemoradiotherapy (nCRT).


DCE-MRI is used to probe the status of tumor microvasculature by means of intravenous contrast injection. In a recent study, DCE-MRI was found to be predictive for pathologic response in patients with esophageal cancer undergoing nCRT 1. Specifically, it was shown that the percentage difference in the 75th percentile AUC between scans acquired before and during nCRT (P75 AUCduring-before) differentiated TRG (tumor regression grade) 1-2 from TRG 3-5 at a threshold of 22.7% (sensitivity 92%, specificity 77%) 1,2. IVIM-MRI also probes the tissue perfusion but with a different mechanism than DCE-MRI. In IVIM-MRI, a high number of diffusion weighted scans is performed and a signal model is fitted to derive 3 parameters: diffusion (D), perfusion fraction (Fp) and pseudo-perfusion (D*) 3. If IVIM parameters are as predictive as DCE-MRI for treatment response, these patients can be spared from intravenous contrast injection. Furthermore, it would allow for repeated response assessment measurements during MRI-guided radiotherapy 4. So far, varying results have been reported for the correlation between the IVIM perfusion-related parameters (D* and Fp) and DCE-MRI parameters outside the esophagus 5,6.


Six patients with biopsy-proven esophageal cancer underwent MRI exams before and after 8-13 fractions of nCRT at a 1.5T system (Philips Ingenia, Best, the Netherlands, see Table 1 for scan parameters). All patients are scheduled to undergo surgery, followed by pathologic assessment of the TRG. A clinician delineated the primary tumor volume on the T2-weighted scans, which was reduced with an isotropic 2 mm margin to account for motion. DCE-MRI studies were analyzed analogous to our previous study 1, including rigid registration to account for breathing motion within the studies, calculation of contrast concentration using T1-mapping and analysis of the AUC (integral of 60 sec, starting after inflow contrast in aorta) to predict outcome for good (TRG 1-2) versus poor (TRG 3-5) response. Coronal IVIM-MRI was acquired with 13 b-values ranging from 0 to 800 s/mm2. Correction for geometric distortions was performed using a B0-field map 7. 3D B-Spline registration on IVIM accounted for motion within scans (registration between even and odd slices) and between the several diffusion-weighted scans using an approach introduced by Guyader et al. 8. IVIM was analyzed performing a bi-exponential fit (non-linear least-squares) 3. Parameters Fp and D were fitted first, using b-values ≥200 s/mm2, followed by fitting of D*. The R2 (goodness of fit) was used to analyze the influence of the B-Spline registration compared to only performing geometric correction. During the analysis of the IVIM parameters a threshold of R2<0.7 was used to exclude voxels with a poor fit. For each patient the median values of IVIM parameters within the delineation were determined for both scans together with the relative difference between the scans.

Results and Discussion

One patient was excluded from analysis due to the small tumor volume (<4mL). The improvement of the registration is clearly visible in Fig. 1, where b0 s/mm2 (transverse view of coronal scans), Fp and the data/fit for one tumor voxel are shown for both approaches. A significant improvement in median R2 values within the delineation was found by registration of the 10 scans (0.72±0.27 [mean±SD] for geometric correction only, 0.93±0.07 for additional registration, p<0.01, Wilcoxon rank test). The AUC and IVIM parameter maps are shown for one scan session (Fig. 2) including the delineation of the tumor. Most parameter maps clearly visualized the tumor volume and comparison of the heterogeneity between AUC and Fp or D* showed an inverse trend. In this preliminary dataset, differences in D* showed no clear relation with response as predicted by the AUC (Fig. 3a). However, Fp against AUC showed an inverse trend for the percentage difference between the two scans per patient (Fig. 3b). Based on the AUCduring-before threshold from our previous study, median Fpduring-before is also able to discriminate the same patients (based on increase/decrease of Fp change). Later, when all patients received surgery, outcome will be verified with pathology.


In five patients with esophageal cancer it was found that IVIM-MRI is feasible and that registration is required to obtain a good fit of the IVIM parameters. Although this study included a limited number of patients, Fp seems the most promising parameter among the perfusion-related IVIM parameters for response prediction. We will continue to include more patients to verify whether IVIM-MRI can replace or possibly complement DCE-MRI for response assessment in esophageal cancer by comparing it with pathological outcome.


No acknowledgement found.


1. Heethuis et al., (2015) Eur Radiol. Submitted 2. Mandard A, Dalibard F, Mandard J (1994) Pathologic assessment of tumor regression after preoperative chemoradiotherapy of esophageal carcinoma. Clinicopathologic correlations. Cancer 2680–2686. 3. Le Bihan D, Breton E, Lallemand D, et al. (1988) Separation of diffusion and perfusion in intravoxel incoherent motion MR imaging. Radiology 168:497–505. 4. Lagendijk JJW, Raaymakers BW, Raaijmakers AJE, et al. (2008) MRI/linac integration. Radiother. Oncol 86(1), 25-29. 5. Jia, Q. J., Zhang, S. X., Chen, W. B., et al. (2014). Initial experience of correlating parameters of intravoxel incoherent motion and dynamic contrast-enhanced magnetic resonance imaging at 3.0 T in nasopharyngeal carcinoma. Eur Radiol, 3076–3087. 6. Yuan, M., Zhang, Y.-D., Zhu, C., et al. (2015). Comparison of intravoxel incoherent motion diffusion-weighted MR imaging with dynamic contrast-enhanced MRI for differentiating lung cancer from benign solitary pulmonary lesions. J. Magn. Reson. Imaging, 00:000–000. 7. Jezzard P, Balaban RS (1995) Correction for geometric distortion in echo planar images from B0 field variations. Magn Reson Med 34:65-73 8. Guyader J-M, Bernardin L, Douglas NHM, et al. (2014) Influence of image registration on apparent diffusion coefficient images computed from free-breathing diffusion MR images of the abdomen. J Magn Reson Imaging 42:315:330.


Table 1: Parameters for acquisition of MRI-protocol scanned before and during treatment.

Figure 1: (a,b) b0 s/mm2 with geometric correction and additional registration, respectively (transverse view). (c,d) Fp is shown for both approaches (coronal view), with one voxel highlighted within delineation (circle). (e,f) The data of the indicated voxel is visible as function of b-value together with the bi-exponential fit.

Figure 2: For the first scan of patient 1 the T2W scan in (a) and DCE-MRI parameter AUC in (b). IVIM parameters D, Fp, D* and goodness of fit R2 are shown in (c-f), respectively. All images include the tumor delineation in blue with the shrunken delineation in red.

Figure 3: For each patient the percentage difference between two scans (with respect to first) visualized for both D* (a) and Fp (b), as function of AUC (with predicted response). D* shows no clear relationship but Fp results in equal differentiation for response prediction (based on sign of Fp).

Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)