In patients with Crohn’s Disease (CD), assessment of fibrosis in stenotic intestinal lesions is critical for therapeutic decisions. Degree of fibrosis may be assessed by region-of-interest (ROI) measurement of magnetic resonance enterography (MRE) metrics such as gadolinium-based contrast gain of enhancement (GE) and magnetization transfer ratio (MTR), which have been associated with the presence of fibrosis in surgically resected lesions^1-3. Parametric mapping (PM) may provide information regarding the extent of fibrotic tissue but can be influenced by respiratory and peristaltic motion. The purpose of this study was to compare various image registration strategies and their impact on quality and robustness of parametric maps of MTR and GE, for assessment of extent of fibrosis in intestinal lesions.

Informed consent was obtained from all subjects involved in this study. Two sets of abdominal MR images were evaluated: (1) a test-retest dataset obtained from healthy individuals, used to assess reliability of parametric maps obtained with different spatial registration algorithms, and (2) a dataset obtained from 7 patients with CD eligible for surgical resection of stenotic lesions.

Test-retest datasets consisted of magnetization transfer contrast (MTC) data acquired at intervals of two weeks using a 3T (GE Discovery) scanner and a dedicated HD Cardiac phased array. MTCs were axial 2D T1-weighted spoiled gradient recalled echo (SGPR) images acquired with and without magnetization transfer prepulse with a 1,100 Hz frequency offset. Scanner settings were: field of view (FOV) = 262-300 mm x 175-200 mm, slice thickness = 5 mm, TR = 30 ms, TE = 4.3 ms, flip angle = 15 degrees, bandwidth = 244 Hz, and image matrix = 192x128x8. Five ROIs were manually defined on the no-prepulse image for the right kidney, right psoas muscle, liver, spinal cord, and phantom that was included in the FOV. All MTC_prepulse images were registered to associated reference MTC_no-prepulse images with multiple spatial registration schemes based on the image processing toolkits FMRIB Software Library (FSL) and Insight Segmentation and Registration Toolkit-based elastix. Linear registrations were performed with elastix and FSL/FLIRT. Nonlinear (Bspline) registrations were performed with elastix and FSL/FNIRT. MTR maps were computed as $$$MTR=100\times(1-{MTC}_{prepulse})/{MTC}_{no-prepulse}$$$. MTR were then normalized to MTR of muscle as follows: $$${MTR}_{norm}=50\times({MTR}/{MTR}_{muscle})$$$. Test-retest reliability was assessed by intra-class correlation (ICC) of MTR_norm in the predefined ROIs.

The patient dataset was a preliminary subset of 7 out of 60 patients with CD enrolled in a multicenter clinical trial involving 6 imaging centers. Data were obtained at 1.5 or 3.0 T and included MTC sequences and fast breath-hold 3D GRE coronal T1-weighted, fat-suppressed series that were acquired prior to IV administered gadolinium-based contrast, approximately 70 seconds (SI_70s, reference) post injection, and 7 minutes (SI_7min) post injection. FSL/FNIRT was used to co-register SI_7min scans to the corresponding SI_70s reference. GE maps were defined as $$$GE=({{SI}_{7min}-{SI}_{70s}})/{{SI}_{70s}}$$$. Cubic ROIs (ROI_fib) of 2.7 x 2.7 x 2.7 cm³ were defined in the region selected for histopathological confirmation of fibrosis. Extent of fibrosis was defined for MTC and GE as the volume within ROI_fib where MTR_norm>30 and GE>0.24 respectively. GE threshold of 0.24 has been reported as predictive of presence of fibrosis independently verified by histopathological examination³.

Generally, spatial registration improved the quality of the parametric maps for MTR (figure 1) and GE (figure 2). MTC test-retest results show that registration significantly improved ICC of ROI-based average MTR_norm. ICC for the original data was 0.83 [0.65 0.91, 95% CI], whereas ICC increased to 0.96 [0.91 0.99] with linear registration, and was 0.90 [0.76 0.96] with the best non-linear algorithm (FSL/FNIRT).

Using the best FSL/FNIRT registration algorithm on contrast enhanced T1-weighted patient data (figure 2) also significantly reduced standard deviation of GE measured within ROI_fib. GE standard deviations for ROI_fib ranged from 0.17 to 0.33 and 0.12 to 0.23, for original and co-registered data respectively. The extent of fibrosis observed from the co-registered data ranged from 0.5 to 5.5 cm³ and 0.8 to 8.8 cm³ within ROI_fib of the GE and MTR_norm maps respectively. Extent of fibrosis measured with GE and MTR_norm were marginally correlated (R=0.66, p=0.1; figure 3).

Spatial registration improved quality and test-retest reliability of MTR and GE parametric maps. In a limited cohort, extents of fibrosis estimated with GE and MTR maps were correlated. Cross validation of MRE-based fibrosis estimates with histological data will be conducted as data becomes available.