Does Feature Compression Limit Hybrid Quantum Medical Image Classification? A Leakage-Controlled Classical-to-Quantum Bottleneck Study

Le résumé fourni par la source

This reproducibility package supports the manuscript: “Does Feature Compression Limit Hybrid Quantum Medical Image Classification? A Leakage-Controlled Classical-to-Quantum Bottleneck Study” The study evaluates the classical-to-quantum feature compression bottleneck in hybrid quantum medical image classification using BreastMNIST and PneumoniaMNIST from MedMNIST v2. A fine-tuned ResNet18 model was used to extract 512-dimensional image features, which were compressed using train-only PCA to 2, 4, 8, and 16 dimensions. Logistic regression, a parameter-matched MLP, and a variational quantum classifier were evaluated under a leakage-controlled protocol using multiple metrics, including accuracy, balanced accuracy, macro-F1, ROC-AUC, PR-AUC, ECE, Brier score, NLL, and training time. This archive contains the reproducibility materials generated during the study, including preprocessed ResNet18 features, PCA-transformed feature files, trained model files/checkpoints, training histories, result tables, leakage-audit outputs, plots, a file manifest, and SHA-256 checksums. The original BreastMNIST and PneumoniaMNIST datasets are publicly available through MedMNIST v2 and are not redistributed in this archive. This package contains derived features, trained models, and experimental outputs generated from those datasets

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.