A Robustness-Oriented Quantum-Classical Hybrid Machine Learning Pipeline for Breast Cancer Diagnosis: External...

Area of Science:

• Quantum Computing
• Machine Learning
• Medical Informatics

Background:

• Breast cancer is a leading cause of mortality.
• Computational decision support complements pathological assessment.
• Quantum classifiers (VQCs, QSVMs) are emerging for medical tasks but require robust validation.

Purpose of the Study:

• To evaluate quantum-classical machine learning pipelines for breast cancer classification.
• To benchmark Variational Quantum Classifiers (VQCs) and Quantum Support Vector Machines (QSVMs) against classical models.
• To develop a rigorous framework for assessing clinical credibility, including calibration, decision-analytic value, and explainability.

Main Methods:

• A quantum-classical pipeline was developed using the Wisconsin Diagnostic Breast Cancer (WDBC) dataset and validated on the Wisconsin Original (WBC) cohort.
• Five classical models (XGBoost, LightGBM, CatBoost, RandomForest, RBF-SVM), two quantum models (VQC, QSVM), and a hybrid ensemble were benchmarked.
• Evaluation included hyperparameter optimization, cross-validation, external validation, ablation studies, noise stress tests, and decision curve analysis.

Main Results:

• Classical RBF-SVM achieved the highest internal discrimination (AUC=0.998).
• On external validation, RBF-SVM, RandomForest, VQC, and the hybrid ensemble showed comparable, near-ceiling discrimination (AUCs ≈ 0.98).
• Embedding rotation significantly impacted VQC performance; noise had a negligible effect. SHAP analysis identified consistent key predictors across models.

Conclusions:

• Quantum machine learning's current strength lies in establishing a rigorous, interpretable benchmarking framework, not superior discrimination over advanced classical models.
• The developed framework allows fair comparison of quantum and classical models.
• Findings represent benchmark-level evidence; further validation is needed for clinical deployment.