This retrospective study was carried out for patients who initially had a tumor stage T3 or greater or N positive rectal cancer and underwent a neoadjuvant CRT with subsequent TME. The study subjects were recruited from a 900-bed tertiary hospital between April 2004 and March 2007. The Institutional Review Board of our institution approved this study, and the need for an informed consent form was waived.

Inclusion criteria were as follows: 1) rectal cancer (≥ cT3 or lymph node involvement) at initial presentation; 2) neoadjuvant CRT; 3) available CT and MRI obtained after neoadjuvant CRT; 4) TME with complete pathological evaluation. The electronic database in the surgery department was searched to identify consecutive patients who underwent surgery for rectal cancer (n = 339). After the review of medical records 269 patients who did not undergo neoadjuvant CRT were excluded. Other patients were excluded because one patient underwent a local excision and 12 patients had no MRI after neoadjuvant CRT.

Digital rectal examinations, abdominopelvic CT and pelvic MRI with or without endorectal ultrasonography were performed for the clinical staging. Patients with rectal cancer and with tumors located within 10 cm of the anal verge were eligible for an inclusion in the neoadjuvant CRT. Patients with a previous or secondary malignancy, with pregnancy a previous large bowel surgery, chemotherapy or with radiotherapy of the pelvis were excluded for the neoadjuvant CRT. Radiation of 45 Gy/25 fractions was delivered to the pelvis, followed by a 5.4 Gy/3 fractions boost to the primary tumor over 6 weeks using linear accelerators with energy of 6 and 15 MV. During the radiation therapy, chemotherapy was performed with one of the following chemotherapy regimens: 1) continuous capecitabine (825 mg/m², two times per day) during the radiation therapy course for 6 weeks (n = 41) or 2) an intravenous bolus injection of 5-fluorouracil (390 mg/m²) plus leucovorin (20 mg/m²) for 3 days in the first and fifth week of radiation (n = 160). About 6 weeks after completion of the neoadjuvant CRT all patients underwent TME, including low anterior resection (n = 54) and abdominoperineal resection (n = 3).

Both, the contrast-enhanced CT and the MRI were performed after the neoadjuvant CRT. The imaging studies were generally scheduled four weeks after the completion of the neoadjuvant CRT to evaluate tumor response and treatment related complications. The surgery was performed two weeks after the imaging studies.

All contrast-enhanced abdominopelvic CT examinations were performed using 16-detector-row CT scanners (Brilliance; Philips Medical Systems, Cleveland, OH, USA). Intravenous nonionic contrast material (2 mL/kg; iopromide, Ultravist 370; Schering, Berlin, Germany) was administered at a rate of 3 mL/s. Bolus tracking software was used to trigger scanning 60 seconds after the aortic enhancement reached a 150 Hounsfield unit threshold. Raw projection data were obtained using the following scanning parameters: scan range, from the diaphragm to upper thigh; detector collimation, 1.5 mm; gantry rotation time, 0.5 seconds; tube potential, 120 kVp; and pitch, 1.17 to 1.25. Effective mAs ranged between 124 and 185 using an automatic tube current modulation technique (Dose-Right, Philips Medical Systems). From the raw data, transverse images were reconstructed: 4 mm thick at 3 mm increments with matrix size of 512 × 512 pixels and field of view of 260-369 mm.

MRI was performed with a 1.5-T system (GyroscanIntera; Philips Medical Systems, Amsterdam, the Netherlands) by using a pelvic phased-array coil. Two-dimensional T1-weighted and T2-weighted fast spin-echo sequences were performed. All sequences were performed in the transverse, coronal and sagittal planes. The transverse planes were angled perpendicular to the long axis of the tumor by using the sagittal plane (11). The imaging parameters for the T1-weighted sequences were a 15 cm field of view, a 4 mm section thickness, a 1 mm intersection gap, 500-582/9.5-13 (repetition time ms/echo time, msec), a 240 × 512 matrix, echo-train length of 4, 3 signals acquired, 90° flip angle, and no fat saturation. The imaging parameters for the T2-weighted sequences were a 15 cm field of view, a 4 mm section thickness, a 1 mm intersection gap, 4200-6069/100-120 (repetition time ms/echo time, msec), a 224 × 512 matrix, echo-train length of 16, 3-4 signals acquired, 90° flip angle, and no fat saturation.

Two board-certified radiologists (Y.H.K. and K.H.L. with 11 and 10 years of experience in gastrointestinal imaging including pelvic MRI, respectively) were recruited for the independent review of CT and MRI obtained after neoadjuvant CRT. They were aware that the patients had T3 or greater or N positive rectal cancer and that the patients received neoadjuvant CRT. The radiologist were blind to the clinical and surgical findings as well as to the histopathology results. All interpretations were conducted on a diagnostic workstation (DS3000, Impax version 4.5; Agfa HealthCare, Mortsel, Belgium) and flat-panel monochrome 3-megapixel monitors (ME315; Totoku, Tokyo, Japan).

Two reading sessions for CT and MRI were held with an interval of four weeks. During the interpretation, CT and MRI obtained before neoadjuvant CRT were also reviewed to refer the initial tumor extent, respectively. They were requested to record T and N stages using the TNM system (12). Because the differentiation between T1 and T2 tumors is known to be virtually impossible on both CT and MRI (13), they were classified into T1/T2 group. A regional node was considered positive if the maximum short axis length was more than 5 mm (14). The distance between the tumor and the potential CRM was measured as the shortest distance (in millimeters) from the outermost part of the tumor to the adjacent mesorectal fascia at the level of the maximum depth of penetration through the rectal wall on transverse images (Fig. 1). The mesorectal fasicia was defined as a linear structure surrounding the mesorectum that shows soft tissue-density on CT and hypo-intensity on T2-weighted MRI. In cases of low rectal cancer where the mesorectal fascia didn't extend, the distance was measured from the tumor to the leavator ani muscle. For MRI, the measurement was conducted on T2-weighted images (Fig. 2). Only the solid portion of the tumor was considered to be a tumor margin, but the fine spiculate border was not applied in the measurement. When a local tumor stage was considered to be T1/T2, the shortest distance to the potential CRM was measured from the bowel wall at the level and site of the tumor. In cases of the tumor invading the mesorectal fascia and leavator ani muscle or other adjacent organs, the distance was zero. When an enlarged lymph node or extramural tumor deposit was located closer to the potential CRM than to the primary tumor, this was used to measure the distance to the CRM (15, 16).

Histopathologic evaluation of the surgical specimen was performed by one gastrointestinal pathologist (H.S.L. with nine years of experience in gastrointestinal pathology) using the method of Quirke et al. (17). After surgery, the specimens were inked to the circumferential resection plane and then fixed in formalin for 24 hours. During the histopathology examination, the local tumor stage was evaluated according to TNM system proposed by the American Joint Committee on Cancer Staging 6th edition (12). The shortest distance between the outmost margin of the tumor and CRM was measured also (17). The histopathology results for the local tumor stage and the distance to CRM served as the reference standard.

The accuracy of T and N stage on CT and MRI were obtained and compared with each other using the McNemar's test. Cases of underestimation and overestimation were counted. The inter-observer agreement for T and N stages was obtained on CT and MRI by weighted kappa statistics.

Sensitivity, specificity, positive predictive value, negative predictive value and accuracy of MRI and CT for the prediction of CRM involvement were obtained and compared with each other using McNemar's test and chi-square test. Cases with the distance to CRM less than or equal to 2 mm were categorized as the involved CRM in both the radiologic and histopathology examinations (18). Inter-observer agreements in predicting the involvement of CRM on both CT and MRI were assessed using weighted kappa statistics. The agreement between measurements for the distance to CRM on both imaging modalities and those on histopathology was assessed with Bland-Altman plots.

In the interpretation of agreement statistics, degrees of agreement were categorized as follows: κ of 0-0.2, slight agreement; κ of 0.2-0.4, fair agreement; κ of 0.4-0.6, moderate agreement; κ of 0.6-0.8, good agreement; and κ of 0.8-1.0, very good agreement (19). p values less than 0.05 were considered statistically significant. All statistical analyses were performed using MedCalc, version 11.6.1 (MedCalc Software, Mariakerke, Belgium).

Accuracies of T and N stages on CT and MRI after neoadjuvant CRT were shown on Table 2. For T stage, the accuracy of MRI was significantly higher than that of CT in both observers (63.2% vs. 43.9%, p = 0.003, for Observer 1; 68.4% vs. 54.4%, p = 0.008 for Observer 2). On both CT and MRI, there was a tendency to overestimate the T stage for both observers. For N stage, the difference of accuracy was not significant between CT and MRI in both observers (77.2% vs. 77.2%, p = 1.00 for Observer 1; 77.2% vs. 80.7%, p = 0.50 for Observer 2). On both CT and MRI, there was a tendency to underestimate the N stage for both observers.

Inter-observer agreements for T stage were good on both CT [weighted κ: 0.69, 95% confidence interval (CI): 0.49 to 0.89] and MRI (weighted κ: 0.74, 95% CI: 0.57 to 0.91). Inter-observer agreements for N stage were very good on both CT (weighted κ: 0.88, 95% CI: 0.72 to 1.0) and MRI (weighted κ: 0.88, 95% CI: 0.72 to 1.0).

An involved CRM was shown in the histopathology of six patients. Both observers identified the potential CRM on CT and MRI in all cases. Diagnostic performances of CT and MRI in predicting CRM involvement were summarized on Table 3. Although a statistically significant difference was not observed, MRI tended to show a better performance than CT. For Observer 1, there was one and zero false negative prediction for CRM involvement on CT and MRI, respectively and also six and five false positive predictions on CT and MRI, respectively. For Observer 2, there were two and zero false negative predictions on CT and MRI, respectively and also six and five false positive predictions on CT and MRI, respectively. The inter-observer agreement in the prediction of tumor involvement of CRM on CT was very good (weighted κ: 0.94, 95% CI: 0.83 to 1.0) and was coincident on MRI (weighted κ: 1.0, 95% CI: 1.0 to 1.0).

Four of the 57 patients with complete tumor regression were excluded from the analysis for the distance to CRM because it was not measurable on histopathology. The Bland-Altman plots showed wide discrepancies between the measurements of the distance to CRM on each CT and MRI and those on histopathology. For Observer 1, the mean difference between the distances measured on CT and histopathology was 1.1 mm (95% CI: 0.2 mm to 2.1 mm) and the limit of agreement was -5.5 mm (95% CI: -7.1 mm to -3.9 mm) to 7.7 (95% CI: 6.1 mm to 9.4 mm) (Fig. 3). For Observer 2, the mean difference between the distances measured on CT and histopathology was 2.2 mm (95% CI: 1.1 mm to 3.4 mm) and the limit of agreement was -6.1 mm (95% CI: -8.1 mm to -4.0 mm) to 10.5 (95% CI: 8.5 mm to 12.5 mm) (Fig. 4). For Observer 1, the mean difference between the distances measured on MRI and histopathology was 1.2 mm (95% CI: 0.5 mm to 1.9 mm) and the limit of agreement was -3.8 mm (95% CI: -5.0 mm to -2.6 mm) to 6.2 (95% CI: 5.0 mm to 7.4 mm) (Fig. 5). For Observer 2, the mean difference between the distances measured on MRI and histopathology was 1.6 mm (95% CI: 0.8 mm to 2.4 mm) and the limit of agreement was -4.2 mm (95% CI: -5.6 mm to -2.8 mm) to 7.3 (95% CI: 5.9 mm to 8.7 mm) (Fig. 6). The inter-observer agreements in measurements of distance to CRM on both CT and MRI were very good [intraclass correlation coefficient (ICC): 0.87, 95% CI: 0.77-0.93 on CT; ICC: 0.95, 95% CI: 0.91-0.97 on MRI].