Development and validation of machine learning-based MRI radiomics models for preoperative lymph node staging in T3 rectal cancer

Abstract

Objective: The present research aimed to evaluate the diagnostic performance of a magnetic resonance imaging (MRI)-based radiomics model for predicting lymph node staging in patients with stage T3 rectal cancer (RC).

Methods: This retrospective study included 225 patients with RC who underwent surgical resection without neoadjuvant therapy treatment. Radiomics features were extracted from high-resolution T2-weighted imaging (T2WI) of primary tumor. Feature selection was performed using the least absolute shrinkage and selection operator (LASSO) algorithm. Five machine learning classifiers were employed to construct radiomics signatures differentiating between N0/N1 (low nodal burden) and N2 (high nodal burden) stages prediction in the training cohort. The predictive performance of each classifier was evaluated using receiver operating characteristic curve analysis, with area under the curve (AUC) comparisons conducted via DeLong's test. Decision curve analysis (DCA) and calibration curves were utilized to assess the clinical utility and calibration performance of the developed models, respectively.

Results: A total of 1,746 radiomics features were extracted from the imaging data, of which 16 features were selected to construct a radiomics signature for lymph node staging in RC. The logistic regression classifier demonstrated the best predictive performance, achieving an AUC of 0.900 [95% confidence interval (CI), 0.848-0.952] in the training cohort. The model's robustness was further validated in the test cohort, with an AUC of 0.876 (95% CI, 0.765-0.986). DCA confirmed the clinical utility of the model.

Conclusions: The radiomics model based on high-resolution T2WI provided an effective and noninvasive approach for preoperatively differentiating between N0/1 and N2 stages in stage T3 RC.

Keywords: MRI; lymph node staging; machine learning; radiomics; rectal cancer.

Figure 1

Flow chart of inclusion and exclusion criteria.

Figure 2

The framework for the radiomics workflow.

Figure 3

(A) Five-fold cross-validation results for alpha selection. The optimal alpha is marked by the dashed line. (B) Coefficient values corresponding to the optimal α value and selected features with non-zero coefficients. (C) MSE deviation on each fold using coordinate descent in five-fold cross-validation.

Figure 4

ROC curves based on five machine learning models in the training and testing cohorts.

Figure 5

(A, B) Boxplots of corresponding radiomics scores in the training and testing cohorts. 0 (blue) represents N0/1 group, 1 (orange) represents N2 group. Asterisk (*) indicates level of statistical significance between categories, with more asterisks representing a higher level of significance. ("***p < 0.001, **p < 0.01, *p < 0.05").

Figure 6

DCA for five models in the training and testing cohorts. (A) DCA of five models in the training cohort. (B) DCA of five models in the testing cohort.