Breast cancer remains the most common neoplasm in women worldwide. Nodal status is a key prognostic factor; therefore, sentinel lymph node biopsy (SLNB) has been established as the gold standard, reducing the need for invasive procedures such as axillary lymphadenectomy. Artificial intelligence (AI) is emerging as a tool that could potentially predict sentinel lymph node positivity and support surgical decision-making. A retrospective cross-sectional observational study was conducted, including 541 patients diagnosed with early invasive breast cancer between 2019 and 2023 at Clínica Universitaria Colombia. Undersampling was applied to balance the dataset. A machine learning and deep learning model was trained using the AIPRIL AI platform. Model performance was evaluated using recall, specificity, accuracy, AUC-ROC, and F1 Score. The selected model was the ridge classifier, which showed an accuracy of 70.28%, precision of 96.46%, recall of 72.02%, specificity of 14%, AUC of 0.49, and F1 score of 82.46%. Variable importance analysis identified initial clinical stage (42.6%), histological grade (22%), and estrogen receptor status (20.4%) as the main predictors. Our model demonstrated good ability to identify positive sentinel lymph node cases, with high accuracy and recall, although with limitations in specificity and AUC. These findings enabled the identification of relevant variables and the acquisition of methodological experience for future research. Furthermore, there is a need to incorporate additional clinically relevant variables to improve the model's performance.
Abi-Saab et al. (Mon,) studied this question.