BACKGROUND: Magnetic resonance imaging (MRI) is the standard diagnostic technique for the locoregional assessment of rectal cancer. The pertinent guidelines recommend neoadjuvant therapy depending on cT and cN categories. In this experimental study, we examined the accuracy of pretreatment MRI staging.

METHODS: MRIs of 50 patients from the OCUM study (non-pretreated, 24 women, histologically confirmed rectal adenocarcinoma, located ≤12cm from the anal verge, stages I-III) were re-assessed by 74 radiologists affiliated with certified colorectal cancer centers. The radiologically determined Union for International Cancer Control (UICC) stages were compared with the histopathological findings, which were unknown to the radiologists. The degree of agreement between the radiologists' assessments and the histopathology was analyzed with κGold, a weighted average of Brennan-Prediger agreement coefficients.

RESULTS: 2915 complete assessments were performed for cTcN (UICC stages); 740 cases (25.4%) were not classifiable because of cTX (0.1%) or cNX (25.3%). The UICC stage accorded with the histopathology in 979 cases (33.6%); there was overstaging in 737 cases (25.3%) and understaging in 459 (15.7%). The κGold for agreement was 0.114 (95% confidence interval, [0.032-0.20]). Agreement was also low for T- and N-categories (κGold=0.354 and κGold=0.235, respectively), but was better for distance to the mesorectal fascia (MRF) (κGold=0.736) and extramural vascular invasion (EMVI) (κGold=0.579).

CONCLUSION: Although MRI remains the standard diagnostic technique for the locoregional assessment of rectal cancer, its accuracy for T- and N-staging in rectal cancer is low, particularly with regard to the distinction of T2 from T3a/b and the assessment of lymph nodes. The assessment of MRF and EMVI is much more reliable.