Quantum‑generated brain MRI data does not improve classifier performance.
Researchers built a testbed that plugs a variational quantum generator into a conditional Wasserstein GAN and compares it with a classical generator of almost the same size (1648 vs. 1632 parameters). Synthetic images were added to a pretrained classifier trained on labeled fractions from 5% to 100% of the data. Eight random seeds and paired significance testing with multiple‑comparison correction showed no variant beating the baseline that uses only real images. The quantum and classical generators also scored the same on diversity and latent‑distribution metrics.
The result matters because earlier papers reported modest accuracy lifts from quantum generators, but those studies often used a single run and mismatched model budgets. This benchmark suggests any low‑data benefit is just regularization, not genuine data expansion, and that quantum generators are not yet a practical shortcut for medical‑image augmentation.
For now, quantum generative models remain a research curiosity rather than a performance edge.