Development and validation of an Interpretable Machine learning model for Discriminating between benign and malignant...

Area of Science:

• Oncology
• Medical Diagnostics
• Biostatistics

Background:

• Early breast cancer detection is crucial for patient prognosis.
• Distinguishing malignant from benign breast lesions often requires invasive procedures like biopsies.
• There is a need for non-invasive methods to support clinical decision-making.

Purpose of the Study:

• To develop and externally validate a predictive model for breast lesion malignancy.
• To utilize routine clinical and laboratory variables for risk stratification.
• To reduce the rate of unnecessary breast biopsies.

Main Methods:

• A retrospective two-center study with development (N=745) and external (N=221) cohorts.
• Logistic regression model development with cross-validation and evaluation on fixed test and external sets.
• Performance metrics included AUC, sensitivity, specificity, F1-score, Brier score, calibration curves, DCA, and SHAP analysis.

Main Results:

• Logistic regression model selected Age, TT, APTT, CEA, and Ca as key predictors.
• Achieved high performance with cross-validated AUC of 0.910 and external validation AUC of 0.861.
• Decision curve analysis demonstrated significant clinical net benefit, and SHAP identified key contributing factors.

Conclusions:

• A logistic model using routine laboratory variables effectively differentiates malignant from benign breast lesions.
• The model shows robust external performance and clear clinical utility for early risk stratification.
• This tool can help minimize unnecessary biopsies and improve breast cancer management.