Do Quantum Kernels Improve Medical Image Classification A Leakage-Controlled Benchmark against Classical RBF-SVM

Résumé fourni par la source

This Zenodo record contains supporting research materials associated with the manuscript “Do Quantum Kernels Improve Medical Image Classification? A Leakage-Controlled Benchmark against Classical RBF-SVM,” accepted for publication in the Journal of Imaging Informatics in Medicine. The study presents a leakage-controlled comparison of quantum support vector machine (QSVM) approaches with established classical machine-learning baselines for medical image classification. Experiments were conducted using the BrainTumorMRI and BreastMNIST datasets. Medical images were represented using frozen ResNet-18 feature embeddings followed by training-only preprocessing and principal component analysis (PCA). Quantum kernel classifiers were evaluated alongside logistic regression, linear SVM, RBF-SVM, and a compact multilayer perceptron under matched experimental conditions, multiple feature dimensions, training budgets, and random seeds. The supporting material provided in this record is intended to facilitate transparency, reproducibility, and independent verification of the analyses reported in the manuscript. It contains derived experimental outputs and related reproducibility information used to generate and validate the results presented in the study. Raw medical-image datasets are not redistributed; these remain available from their respective original repositories and are subject to the corresponding dataset licenses and terms of use. The results of the study show that, within the evaluated leakage-controlled settings, the tested quantum kernel methods did not outperform strong classical baselines. The accompanying materials are provided to support further benchmarking and research on rigorous classical–quantum comparisons in medical image classification.