Clinical Report: Integration of Multimodal Features for Risk Assessment of Malignancy in Breast Masses
Overview
This study developed machine learning models that integrate multimodal features to enhance breast mass malignancy risk stratification. The Random Forest model using combined features demonstrated the highest diagnostic performance, particularly in lower-risk BI-RADS categories.
Background
Breast cancer is the most commonly diagnosed cancer in women, with significant morbidity and mortality rates. Accurate risk stratification of breast masses is crucial for timely intervention and reducing unnecessary procedures. Traditional diagnostic methods have limitations, highlighting the need for advanced technologies like machine learning to improve diagnostic accuracy.
Data Highlights
Model
AUC
95% CI
Random Forest (combined features)
0.850
0.810-0.875
Logistic Regression (BI-RADS)
0.820
0.775-0.856
Random Forest (ultrasound)
0.800
0.768-0.839
Logistic Regression (radiomics)
0.740
0.706-0.780
Key Findings
The Random Forest model achieved the highest overall performance with an AUC of 0.850.
Logistic Regression performed best with BI-RADS terminology features (AUC 0.820).
Subgroup analysis showed excellent performance for BI-RADS categories 2 (AUC 1.000) and 3 (AUC 0.947).
Performance was poor for higher-risk categories 4b (AUC 0.649) and 4c (AUC 0.551).
Machine learning models can reduce unnecessary biopsies and improve diagnostic confidence.
Clinical Implications
The integration of multimodal features in machine learning models can enhance the accuracy of breast mass malignancy risk assessment. Clinicians may consider these models to improve decision-making and reduce the number of unnecessary biopsies, particularly in lower-risk categories.
Conclusion
The study highlights the potential of machine learning in breast cancer diagnostics, particularly in improving risk stratification. Further refinement and validation of these models are necessary before widespread clinical implementation.
