A novel approach combining radiomics analysis of ultrasound (US) and magnetic resonance imaging (MRI) data significantly improves the prediction of disease-free survival (DFS) in patients with breast cancer. The study, published in Breast Cancer Research, integrates multi-modal radiomics signatures with clinical and traditional MRI features to enhance prognostic accuracy.
The study, conducted at Sun Yat-sen University Cancer Center (SYUCC) and other hospitals, involved 643 patients who underwent preoperative US and MRI. Researchers developed automatic segmentation-based radiomics signatures (ASFs) for both US and MRI, focusing on both intra-tumoral and peri-tumoral regions of interest (ROIs). These signatures were then combined to create a gross-radiomics signature, which was validated across training, internal, and external testing sets.
Radiomics Signature Construction and Validation
Radiomics features were extracted using the PyRadiomics package, adhering to Image Biomarker Standardization Initiative (IBSI) guidelines. The least absolute shrinkage and selection operator (LASSO) Cox regression model was used to select features for constructing intra- and peri-radiomics signatures. The resulting Rad-score, calculated for each patient, was used to divide patients into low- and high-risk groups based on an optimal cutoff identified by X-tile analysis.
Kaplan-Meier survival analysis and log-rank tests were used to evaluate survival rates between the risk groups. Harrell’s concordance index (C-index) and time-dependent receiver operating characteristic (ROC) curves assessed the performance of the radiomics signature in predicting DFS. The DeLong test was used to compare the area under the ROC curves (AUC) to assess discriminatory power.
Model Performance and Clinical Integration
Six predictive models were constructed using Cox regression, including a clinical model, a traditional MRI model, a clinical traditional MRI model, a multi-modal radiomics signature model, a multi-modal radiomics model, and a multi-modal clinical imaging model. The multi-modal clinical imaging model, which combined the multi-modal Rad-score with the clinical traditional MRI model-score, demonstrated the best performance.
The models were evaluated based on C-index, calibration curves, decision curve analysis (DCA), and Bayesian information criterion (BIC) values. The multi-modal clinical imaging model showed superior discrimination and clinical usefulness compared to other models.
Clinical Implications
The integration of radiomics with standard imaging and clinical data offers a more comprehensive approach to predicting breast cancer prognosis. By incorporating both intra-tumoral and peri-tumoral features from US and MRI, the multi-modal radiomics signature captures a broader range of tumor characteristics that influence disease progression. This approach has the potential to improve patient stratification and treatment planning, ultimately leading to better outcomes for women with breast cancer.
"The findings suggest that multi-modal radiomics analysis can provide valuable insights into tumor biology and improve the accuracy of prognostic predictions," the authors noted. "Further research is needed to validate these findings in larger, multi-center studies and to explore the potential of radiomics to guide personalized treatment strategies."
