Predictive value of MRI-detected tumor deposits in locally advanced rectal cancer

Although tumor deposits (TDs) are not the same as lymph nodes, the prognosis of patients with TDs is similar or worse than that of patients with metastatic lymph nodes. TDs are mostly assessed by the histology of samples after surgery, thus, not helpful for preoperative treatment strategies. The primary objective of this study was to detect TDs by MRI and evaluate its predictive value.

A total of 114 patients with rectal cancer were retrospectively analyzed. Clinicopathological and MRI data mainly including MRI- detected TDs (mTDs), tumor border configuration (TBC) on MRI, MRI-detected extramural vascular invasion (mEMVI), MRI-detected lymph node metastasis (mLN), MRI T stage, MRI N stage, the range of rectal wall involved by the tumor, peritoneal reflection invasion, tumor length, tumor location, cord sign at the tumor edge, nodular protrusion at the tumor edge, maximal extramural depth and pathology-proven lymph node involvement (pLN) were evaluated. The correlation of MRI factors with postoperative distant metastasis (PDM) and pLN were analyzed by univariate analysis and multivariate logistic regression analysis, and nomograms were established based on the latter. The diagnostic efficiency was evaluated by the receiver operating characteristic curve (ROC) and area under the curve (AUC).

A total of 38 cases of pLN, 13 of PDM and 17 of pathology-proven TDs (pTDs) were found. Ten cases of PDM and 22 cases of pLN in 30 mTDs cases were also found. Chi-square test showed that mTDs, mLN, TBC, mEMVI, MRI T stage, nodular protrusion, cord sign, maximal extramural depth and peritoneal reflection invasion were correlated with PDM and pLN (P<0.05). mTDs and peritoneal reflection invasion were independent risk factors for PDM (odds ratio: 10.15 and 8.77, P<0.05), mTDs and mLN were independent risk factors for pLN (odds ratio: 5.50 and 5.91, P<0.05), and Hosmer-Lemeshow test showed that the results of two models were not statistically significant, suggesting that the fit was good. On this basis, two nomograms for predicting PDM and pLN were confirmed by Bootstrap self-sampling, and the C-indices of the two nomograms were 0.837 and 0.817, respectively. The calibration curves and ROC curves of the two nomograms showed that the correlation between the predicted and the actual incidence of PDM and pLN was good. The DeLong test showed that the predictive efficiency of the nomogram in predicting pLN was better than that of mLN (P=0.0129).

mTDs are a risk factor for PDM and lymph node metastasis. The two nomograms based on mTDs showed a good performance in predicting PDM and lymph node metastasis, possessing a certain clinical value.