Over-the-Counter Breast Cancer Classification Using Machine Learning and Patient Registration Records.

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning

Background:

• Breast cancer risk estimation is crucial for early detection and intervention.
• Traditional screening methods may have limitations in accessibility and early risk stratification.
• Leveraging readily available patient data for risk assessment is an area of active research.

Purpose of the Study:

• To assess the feasibility of using machine learning (ML) models with patient registration data for an over-the-counter (OTC) breast cancer risk screening tool.
• To identify the most effective ML model for breast cancer risk estimation using limited demographic and clinical information.
• To evaluate the performance of the developed model across different mammographic density groups.

Main Methods:

• Retrospective collection of data from women presenting with breast-related issues at Hospital Universiti Sains Malaysia.
• Development and comparison of eight ML models: k-nearest neighbour (kNN), elastic-net logistic regression, MARS, ANN, PLS, random forest, SVM, and extreme gradient boosting.
• Feature importance assessment using a model-agnostic approach, focusing on data available in patient registration forms.

Main Results:

• The k-nearest neighbour (kNN) model achieved the highest Youden J index, precision, and PR-AUC.
• The support vector machine (SVM) model recorded the highest F2 score.
• The selected kNN model demonstrated balanced sensitivity, specificity, and PR-AUC across various mammographic density levels, with age at examination identified as the most critical feature.

Conclusions:

• Machine learning models utilizing patient registration information are feasible for developing an OTC breast cancer risk screening tool.
• The kNN model shows promise for accessible breast cancer risk assessment.
• Further validation and implementation of such models could enhance early breast cancer detection strategies.