Patients
This retrospective study was approved by the institutional review board of the First Affiliated Hospital, Sun Yat-sen University with waiver of informed consent.
A retrospective search of the databases of a tertiary referral institution was performed to retrieve all consecutive patients diagnosed with rectal neoplasms who underwent high-resolution rectal MRI examination in our hospital from January 2015 to December 2016 were recruited. Among them, patients who met the following criteria were excluded: (1) carcinoma of the anal canal or sigmoid; (2) lack of histological results, or histologically proven adenoma, gastrointestinal stromal tumour (GIST), mucinous or composition of mucous adenocarcinoma; (3) recurrent rectal carcinoma; (4) underwent neoadjuvant chemoradiotherapy before MRI examination; (5) lack of IVIM DWI sequences; (6) artifacts producing poor image quality; and (7) insufficient space for placing region of interest (ROI) in the mesorectum.
MRI protocol
All patients were routinely intramuscularly injected with 20 mg of anisodamine to minimize intestinal peristaltic movement 10 min before the examination. Gadopentetate dimeglumine (0.2 mL/kg body weight) was intravenously injected using a power injector at a rate of 3.0 mL/s followed by a 25 mL saline flush.
The rectal MRI examination (Table 1) was performed using a 3 T MRI scanner (Magnetom Verio; Siemens Healthcare) equipped with a six-channel body-matrix coil. Routine rectal MRI included (1) axial, sagittal, coronal and oblique axial (performed perpendicular to the long axis of the rectum at the level of the tumour) T2-weighted imaging and (2) contrast-enhanced fat-saturated oblique axial and 3D coronal T1-weighted imaging. The IVIM DWI sequence was performed prior to gadolinium injection. A total of 14 b values (0, 5, 10, 20, 30, 40, 60, 80, 100, 150, 200, 400, 600 and 1000 s/mm2) were applied using a single-shot spin-echo echo-planar-imaging sequence.
Image interpretation
A computerized radiologic database was used for patient selection and image interpretation. Two consultant radiologists, with more than 5 years of specialized training in rectal MRI independently reviewed the images. Once image interpretation was completed, any disagreement between the observers was resolved by a third investigator (with 25 years of experience in radiology of gastrointestinal tract) and the majority view was taken as the final consensus.
The mesorectal fascia (MRF) is easily identified on axial T2-weighted images as a thin hypointense line. Based on the relationship between the tumour and the MRF, definitive or potentially positive mrCRM involvement was defined as tumour infiltration, suspicious lymph nodes, extramural vascular invasion or disseminated lesions within 1 mm of the MRF [19].
EMVI refers to the invasion of tumour cells into the small vessels outside the rectal wall. mrEMVI was identified as a serpiginous extension of the tumour signal within the vascular structure (in which a vessel was defined as a tubular structure that contained a signal void on T2-weighted images and was continuous across adjacent slices), including a slightly expanded contour and caliber and an obvious irregular vessel contour or nodular expansion of the involved vessels [20].
IVIM DWI analysis
The data set was analyzed on the basis of the bi-exponential IVIM DWI model introduced by Le Bihan [21] using the following equation: Sb/ S0 = (1 − f) exp (-b·D) + f·exp [ (-b·(D*)], where Sb is the signal intensity at the particular b-value used, S0 is the signal intensity at the b-value of 0, f is the fraction of the signal linked to microcirculation, D is the true diffusion coefficient representing molecular diffusion of pure water, and D* represents perfusion-related incoherent microcirculation, and is known as the pseudodiffusion coefficient.
The IVIM DWI data were postprocessed using a software developed in house (MATLAB Version 3.1; Mathworks, Natick, MA, USA), and the parameter maps of IVIM DWI (D, D* and f parameter maps) were automatically generated in a voxel-by-voxel manner using all 14 b values. A segmented fitting method was used for a more robust calculation, similar to the implementation in a previous study [22]. D and f were first estimated by assuming that the perfusion fraction of the signal could be neglected at a b-value > 200 s/mm2; then, D* was calculated by applying the previously calculated D and f to the IVIM model.
All ROIs were manually delineated and contoured, and they were confirmed by comparing the position in the DWI data (b = 0 s/mm2) or D-mapping to that in the axial T2-weighted imaging. The ROIs were placed in the mesorectum adjacent to (MAT, within 5 mm of the outline of the largest cross-section of the tumour or in the upper or lower slices closest to the tumour if there was no available space) and distant from the tumour (MDT, mesorectum contralateral to the semiperimeter tumour cross-section or at least 10 mm from the full-perimeter tumour cross-section) in each case. The sizes of the two ROIs were kept as equal as possible in the same case. Moreover, the ROIs of both the MAT and MDT were drawn to avoid the tumour infiltration area, the invaded vasculature, and suspicious lymph node or dissemination (Fig. 1). The mean value of each IVIM parameter within a ROI on the IVIM parameter maps was recorded for analysis. The mean size of the ROIs of MAT and MDT was 21.12 ± 7.44 mm2 and 18.02 ± 7.23 mm2, respectively. Additionally, IVIM parameters were processed independently by two radiologists to assess interobserver reproducibility.
Statistical analysis
Statistical analyse were performed using SPSS v20.0 (IBM Inc., Armonk, NY, USA) and MedCalc Statistical Software v15.8 (MedCalc Software bvba, Ostend, Belgium). All measurements of IVIM parameters were presented using medians (inter-quartile range, IQR). Data distribution normality was assessed using the Kolmogorov–Smirnov or Shapiro–Wilk test. Two independent-sample t or Mann–Whitney U tests were used to analyze the differences in the IVIM parameters of the MAT and MDT between the mrCRM-positive/ negative and mrEMVI-positive/ negative groups, respectively. A paired-samples t or Wilcoxon test was used to analyze the differences between the MAT and MDT for each mrCRM and mrEMVI group. The interobserver agreement was assessed with Cohen’s kappa (κ) for evaluating mrCRM and mrEMVI status. κ values of 0–0.20 are characterized as poor agreement, 0.21–0.40 as fair, 0.41–0.60 as moderate, 0.61–0.80 as good, and 0.81–1.00 as excellent. The intraclass correlation coefficient (ICC) was used to assess the interobserver agreement for the IVIM parameters. ICCs > 0.75, 0.5–0.75, and < 0.5 were considered good, moderate and poor agreement, respectively. Moreover, Bland–Altman plots were constructed and limits of agreement (LoAs) based on the plots were estimated to evaluate IVIM measurement reproducibility. A two-tailed p value < 0.05 indicated statistical significance.
