Abstract
Objective
To establish and validate radiomics-based machine learning models based on dynamic contrast–enhanced magnetic resonance imaging (DCE-MRI) for the preoperative identification of sentinel lymph node metastases (SLNM) in patients with clinical N0 (cN0) breast cancer.
Methods
Preoperative DCE-MRI images of patients with cN0 breast cancer were collected from September 2006 through December 2021 from 144 SLNM-positive patients and 144 age-matched SLNM-negative patients. The patients were randomly divided into training (n = 200) and validation (n = 88) sets. Radiomic features were extracted from the first phase of the DCE-MRI. The least absolute shrinkage and selection operator (LASSO) algorithm was used to select the radiomics features. Four machine learning classifiers were evaluated: k-nearest neighbor, random forest, support vector machine, and eXtreme Gradient Boosting.
Results
Five radiomic features were selected using LASSO logistic regression. Our radiomics models showed good calibration and prediction values with areas under the receiver operating characteristic curve from 0.70 to 0.77 and from 0.68 to 0.75 in the training and validation sets, respectively. In the validation set, the SVM model achieved the highest value with an AUC of 0.75, with a sensitivity of 70.5%, specificity of 77.3%, and accuracy of 73.9%.
Conclusion
MRI radiomics-based machine learning models can be useful for preoperative prediction of SLNM in cN0 breast cancer.