Related Experiment Video For Shapley additive explanations
Updated: Jun 9, 2026
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
Multi-parametric MRI habitat radiomics with interpretable machine learning for early prediction of axillary lymph
Bo Xie1,2, Xue Peng1,2, Yueyan Wang3
1The First Affiliated Hospital of Bengbu Medical University, Bengbu, China.
Background:
Accurate preoperative assessment of axillary lymph node metastasis (ALNM) in triple-negative breast cancer (TNBC) is essential for treatment planning. Conventional radiomics may overlook intratumoral heterogeneity (ITH) and lacks interpretability.
Purpose:
The study aimed to develop and validate an interpretable, multiparametric magnetic resonance imaging (mpMRI)-based habitat radiomics model for the preoperative prediction of ALNM in TNBC.
Materials And Methods:
In this retrospective study, 125 patients with pathologically confirmed TNBC who underwent preoperative mpMRI were included. Tumors were manually segmented, and an unsupervised clustering approach was used to partition each lesion into intratumoral habitats. Radiomics features were extracted from both the whole tumor and habitat subregions. Clinical, conventional radiomics, habitat radiomics, and combined models were constructed using the eXtreme Gradient Boosting (XGBoost) algorithm. Model performance was evaluated using receiver operating characteristic (ROC) curve analysis, calibration curves, and decision curve analysis (DCA). Model discrimination was compared using DeLong's test. Shapley additive explanations (SHAP) were used for model interpretation.
Results:
In the training set, the clinical, conventional radiomics, habitat radiomics, and combined models achieved AUCs of 0.68, 0.76, 0.79, and 0.82, respectively. In the test set, the corresponding AUCs were 0.66, 0.70, 0.74, and 0.81. The combined model showed the best performance, while the habitat radiomics model outperformed the conventional radiomics and clinical models. In the test set, SHAP analysis identified axillary lymph node (ALN) length as the most important predictor.
Conclusion:
The habitat radiomics model improved predictive performance over the conventional radiomics model for preoperative ALNM assessment in TNBC, and the combined model showed the highest performance. This interpretable framework highlights the value of intratumoral heterogeneity characterization for non-invasive nodal risk stratification.
