Related Experiment Video
Updated: Aug 12, 2026
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
Development and External Validation of an Interpretable Machine Learning Model Using Routine Blood Biomarkers for
Longmei Chen1, Xiaoyan Teng2, Jiale Tian3
1Department of Laboratory Medicine, Shanghai Baoshan Hospital of Integrated Traditional Chinese and Western Medicine (Baoshan Hospital, Shanghai University of Traditional Chinese Medicine), Shanghai, 201999, People's Republic of China.
This study developed an interpretable machine learning model using routine blood biomarkers to identify malignant breast nodules. The model shows promise as a complementary tool for risk assessment alongside imaging, aiding in precise clinical intervention.
Area of Science:
- Biomedical Engineering
- Computational Biology
- Oncology
Background:
- Accurate identification of malignant breast nodules is crucial for timely clinical intervention.
- Routine blood biomarkers offer a potential non-invasive method for breast nodule assessment.
Purpose of the Study:
- To develop and externally validate an interpretable machine learning model for differentiating malignant from benign breast nodules using routine blood biomarkers.
- To assess the model's performance and interpretability in independent validation cohorts.
Main Methods:
- A retrospective multicenter study involving 899 women with pathologically confirmed breast nodules.
- Development of eight machine learning models, with Random Forest (RF) selected for its performance, using routine blood biomarkers and age.
- External validation in temporal and independent cohorts, with performance evaluated by AUC, calibration curves, and decision curve analysis (DCA). SHapley Additive exPlanations (SHAP) were used for interpretability.
Main Results:
- The RF model achieved AUCs of 0.82 (internal), 0.78 (temporal), and 0.72 (external validation), demonstrating acceptable discriminative ability.
- The model's diagnostic performance was comparable to ultrasound and mammography in the temporal validation cohort.
- SHAP analysis identified high-sensitivity C-reactive protein (hs-CRP), red blood cell count (RBC), and age as key predictors.
Conclusions:
- Interpretable machine learning models utilizing routine blood biomarkers can effectively support the differentiation of malignant from benign breast nodules.
- The developed model can serve as a complementary pre-biopsy risk assessment tool alongside conventional imaging.
- Further prospective validation in diverse populations and healthcare settings is recommended before routine clinical implementation.