Can Quantum Encoding Choice Close the Classical–Quantum Performance Gap in Medical Image Classification? A Controlled Five-Encoding Ablation

Résumé fourni par la source

This repository contains the supplementary material and reproducibility package for the manuscript “Can Quantum Encoding Choice Close the Classical–Quantum Performance Gap in Medical Image Classification? A Controlled Five-Encoding Ablation.” The study evaluates five quantum encoding strategies for variational quantum classifiers in a hybrid classical–quantum medical image classification pipeline. The experiments were conducted on BreastMNIST and PneumoniaMNIST from the MedMNIST v2 benchmark. The package includes result tables, mean–standard deviation summaries across three random seeds, generated figures, training logs, processed feature files, duplicate-check information, source code, notebooks, configuration details, README files, and checksum/manifest files. The experimental pipeline includes leakage-controlled preprocessing, ResNet18 fine-tuning, train-only standardization and PCA-4 compression, tanh-based quantum range mapping, comparison of logistic regression, matched-parameter MLP, and five VQC encodings: RY angle encoding, RX/RY/RZ angle encoding, data re-uploading, ZZ feature map, and Pauli feature map. This package is intended to support reproducibility, transparency, and verification of the reported classification, calibration, seed-robustness, and computational-efficiency results. Raw BreastMNIST and PneumoniaMNIST data are publicly available from the MedMNIST v2 benchmark; this repository provides the processed outputs and analysis materials generated for the study.

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