Prediction of Lymph Node Metastasis In Rectal Adenocarcinoma Based on Multi-Sequence MRI Radiomics with Clinical Features: A Two-Center Study
Abstract
Purpose
To establish and validate a combined model based on multi-sequence MRI radiomics and clinical features for predicting LN metastasis in rectal adenocarcinoma patients.
Methods
We conducted a retrospective study on 160 rectal adenocarcinoma patients. The status of rectal adenocarcinoma was confirmed by histopathology. 120 patients from Hospital I were divided into the training set (83 cases) and internal validation set (37 cases). 40 patients from Hospital II were served as the external testing set. Each patient received DWI and automatically generated the corresponding ADC, CE-T1WI and T2WI sequence examination. For each MRI sequence, the three-dimensional volume of interest (VOI) was delineated based on primary tumor and a total of 851 radiomics features were extracted. Spearman's rank correlation coefficient and the least absolute shrinkage and selection operator (LASSO) were used for feature selection. One-way analysis of variance (ANOVA) used to determine the risk factors for LN metastasis in rectal adenocarcinoma patients.
Results
One-way ANOVA showed that only one clinical features, tumor differentiation, was closely related to LN metastasis in rectal adenocarcinoma (p<0.05). The combined model performed well both in the internal validation set and external testing set with AUC of 0.868 (95%CI, 0.750-0.985) and 0.859 (95%CI, 0.750-0.977). Its sensitivity, accuracy and specificity were the highest compared to other five models, with values of 0.845, 0.837, 0.846 and 0.839, 0.819, 0.821 in validation set and external testing set, respectively.
Conclusion
Tumor differentiation is an important factor in LN metastasis of rectal adenocarcinoma. The model combining radiomics features of primary tumor volume based on multi-sequence MRI with clinical features had good predictive performance in LN metastasis of rectal adenocarcinoma.