High-resolution MRI-based radiomics analysis to predict lymph node metastasis and tumor deposits respectively in rectal cancer

Purpose To establish and validate two predictive radiomics models for preoperative prediction of lymph node metastases (LNMs) and tumor deposits (TDs) respectively in rectal cancer (RC) patients. Methods A total of 139 RC patients (98 in the training cohort and 41 in the validation cohort) were enrolled in the present study. High-resolution magnetic resonance images (HRMRI) were retrieved for tumor segmentation and feature extraction. HRMRI findings of RC were assessed by three experienced radiologists. Two radiomics nomograms were established by integrating the clinical risk factors, HRMRI findings and radiomics signature. Results The predictive nomogram of LNMs showed good predictive performance (area under the curve [AUC], 0.90; 95% confidence interval [CI] 0.83-0.96) which was better than clinico-radiological (AUC, 0.83; 95% CI 0.74-0.93; Delong test,p = 0.017) or radiomics signature-only model (AUC, 0.77; 95% CI 0.67-0.86; Delong test,p = 0.003) in training cohort. Application of the nomogram in the validation cohort still exhibited good performance (AUC, 0.87; 95% CI 0.76-0.98). The accuracy, sensitivity and specificity of the combined model in predicting LNMs was 0.86,0.79 and 0.91 in training cohort and 0.83,0.85 and 0.82 in validation cohort. As for TDs, the predictive efficacy of the nomogram (AUC, 0.82; 95% CI 0.71-0.93) was not significantly higher than radiomics signature-only model (AUC, 0.80; 95% CI 0.69-0.92; Delong test,p= 0.71). Radiomics signature-only model was adopted to predict TDs with accuracy=0.76, sensitivity=0.72 and specificity=0.94 in training cohort and 0.68, 0.62 and 0.97 in validation cohort. Conclusion HRMRI-based radiomics models could be helpful for the prediction of LNMs and TDs preoperatively in RC patients.

Automated summary

Two radiomics models were established for predicting lymph node metastases and tumor deposits in rectal cancer patients preoperatively, showing good performance in both training and validation cohorts.