Do Hybrid Classical–Quantum Models Offer a Reliable Advantage in Medical Image Classification? A Leakage-Controlled and Explainability-Aware Evaluation

This dataset contains the verified supplementary results supporting the manuscript “Do Hybrid Classical–Quantum Models Offer a Reliable Advantage in Medical Image Classification? A Leakage-Controlled and Explainability-Aware Evaluation.” The study evaluates hybrid classical–quantum machine-learning approaches for medical image classification using BreastMNIST, PneumoniaMNIST, and BrainTumorMRI datasets. Classical baselines and variational quantum classifiers are compared under leakage-controlled and reproducible experimental settings. The deposited material includes final multi-seed experimental results, prediction-level outputs, classification reports, confusion matrices, paired bootstrap results, McNemar tests with Holm correction, duplicate and near-duplicate leakage audits, PCA and quantum-input mapping diagnostics, 12-seed robustness results, the BrainTumorMRI matched feasibility control, cross-seed explainability stability results, deletion-based sensitivity analysis, and SHAP feature-importance outputs. The archive contains derived experimental results only and does not redistribute the original medical-image datasets. BreastMNIST and PneumoniaMNIST remain available through MedMNIST v2, while BrainTumorMRI remains available from its original Kaggle source. This supplementary dataset is provided to support transparency, reproducibility, and independent verification of the results reported in the associated manuscript.