QML-MedImage tops classical SVMs on X-rays

Interesting result, but it’s still a tightly controlled benchmark rather than proof that quantum kernels beat classical methods in the wild.

• –The strongest signal is not raw accuracy, but minority-class recall: the classical linear kernel repeatedly collapses to majority-class prediction while QSVM keeps useful recall.
• –The setup is unusually careful for a quantum ML paper, since both sides get the same PCA-q features, which makes the comparison more defensible.
• –The claim is still bounded by noiseless simulation and a narrow binary task, so the next test is whether the advantage survives hardware noise and larger-scale data.
• –The paper is most compelling as a feature-geometry story: on these embeddings, kernel choice matters more than the usual “classical vs quantum” framing suggests.
• –If the codebase is usable, the repo could matter as a reproducible benchmark for anyone studying where low-dimensional medical embeddings become linearly brittle.