OBJECTIVE: To determine the value of radiomics in identifying lymph node (LN) metastasis in patients with rectal nonmucinous adenocarcinoma. Methods: Imaging data of 91 patients were retrospectively analyzed (61 in the training set and 30 in the test set). A total of 1 301 radiomics features were extracted from high-resolution T2-weighted images of the whole primary tumor. The least absolute shrinkage and selection operator (LASSO) logistic regression was performed to choose the optimal features and construct a radiomics classifier in the training set. Its discrimination performance was compared with that of morphological criteria by receiver operating characteristic (ROC) curve analysis, which was validated in the test set. Results: The radiomics classifier combined with five key features was significantly associated with LN metastasis, which distinguished LN metastasis with an area under curve (AUC) at 0.874 (95% CI 0.787 to 0.960) in the training set, and the performance was similar in the test set (AUC 0.878, 95% CI 0.727 to 1.000). The AUCs according to the morphological criteria in the training set and test set were 0.619 (95% CI 0.487 to 0.752) and 0.556 (95% CI 0.355 to 0.756), respectively. Discrimination of the radiomics classifier was superior to that of morphological criteria in both the two datasets (both P <0.05). Conclusion: The radiomics classifier provides individualized risk estimation for LN metastasis in rectal nonmucinous adenocarcinoma patients and it has the advantage over the morphological criteria.
OBJECTIVE: To determine the value of radiomics in identifying lymph node (LN) metastasis in patients with rectal nonmucinous adenocarcinoma. Methods: Imaging data of 91 patients were retrospectively analyzed (61 in the training set and 30 in the test set). A total of 1 301 radiomics features were extracted from high-resolution T2-weighted images of the whole primary tumor. The least absolute shrinkage and selection operator (LASSO) logistic regression was performed to choose the optimal features and construct a radiomics classifier in the training set. Its discrimination performance was compared with that of morphological criteria by receiver operating characteristic (ROC) curve analysis, which was validated in the test set. Results: The radiomics classifier combined with five key features was significantly associated with LN metastasis, which distinguished LN metastasis with an area under curve (AUC) at 0.874 (95% CI 0.787 to 0.960) in the training set, and the performance was similar in the test set (AUC 0.878, 95% CI 0.727 to 1.000). The AUCs according to the morphological criteria in the training set and test set were 0.619 (95% CI 0.487 to 0.752) and 0.556 (95% CI 0.355 to 0.756), respectively. Discrimination of the radiomics classifier was superior to that of morphological criteria in both the two datasets (both P <0.05). Conclusion: The radiomics classifier provides individualized risk estimation for LN metastasis in rectal nonmucinous adenocarcinomapatients and it has the advantage over the morphological criteria.