Multimodal feature fusion model for breast mass malignant risk stratification
-
By
-
Shengxin Pei
-
Xiumei Tang
-
Hongxia Su
-
Jingyan Liu
-
Zihan Lan
-
Siyu Wang
-
Yulan Peng
-
June 3, 2026
-
Clinical Scorecard: Integration of Multimodal Features for Risk Assessment of Malignancy in Breast Masses
At a Glance
| Category | Detail |
|---|
| Condition | Breast Mass Malignancy |
| Key Mechanisms | Integration of BI-RADS terminology, ultrasound imaging, and radiomics features. |
| Target Population | Women with breast masses undergoing ultrasound imaging. |
| Care Setting | Single medical center, retrospective cohort study. |
Key Highlights
- Random Forest model achieved the highest AUC of 0.850 for malignancy risk stratification.
- Logistic Regression performed best with BI-RADS features (AUC 0.820).
- Subgroup analysis showed excellent performance for BI-RADS categories 2 and 3.
- Performance was poor for higher-risk categories 4b and 4c.
- Machine learning models can potentially reduce unnecessary biopsies.
Guideline-Based Recommendations
Diagnosis
- Utilize machine learning models integrating multimodal features for improved risk stratification.
Management
- Consider the use of ultrasound imaging and radiomics in conjunction with BI-RADS terminology.
Monitoring & Follow-up
- Evaluate model performance across different BI-RADS categories for ongoing refinement.
Risks
- Acknowledge limitations in performance for higher-risk BI-RADS categories.
Patient & Prescribing Data
Women with benign and malignant breast masses.
Machine learning models may assist in decision-making regarding biopsies.
Clinical Best Practices
- Incorporate multimodal features in breast mass evaluation.
- Regularly update models based on new data for improved accuracy.
- Conduct multicenter validation studies to enhance generalizability.
Related Resources & Content
