BUSI
A public breast ultrasound image dataset containing benign, malignant, and normal samples. It is commonly used in breast cancer detection and medical image classification research and exhibits notable class imbalance.
Timeline (last 90 days)
Researchers used quantum generative models to generate synthetic images on a breast ultrasound image dataset (BUSI) and trained classifiers with these synthetic data, ultimately achieving performance on the real test set that was better than a pure real-data baseline.
The classifier trained entirely on synthetic images achieved an AUC-ROC of 0.802 on the real test set, while the baseline trained only on real data achieved 0.760.
The research team designed three training schemes for comparison: training using only real images, training using only synthetic images, and mixed training with real and synthetic images.
The paper does not discuss the comparison between the training cost of quantum generative models and traditional GANs, and the BUSI dataset is relatively limited in size, so whether this method can generalize to larger-scale image datasets with more cancer types remains unknown.
All Facts (4)
The classifier trained entirely on synthetic images achieved an AUC-ROC of 0.802 on the real test set, while the baseline trained only on real data achieved 0.760.
50%UnverifiedThe research team designed three training schemes for comparison: training using only real images, training using only synthetic images, and mixed training with real and synthetic images.
50%UnverifiedThe paper does not discuss the comparison between the training cost of quantum generative models and traditional GANs, and the BUSI dataset is relatively limited in size, so whether this method can generalize to larger-scale image datasets with more cancer types remains unknown.
50%UnverifiedResearchers used quantum generative models to generate synthetic images on a breast ultrasound image dataset (BUSI) and trained classifiers with these synthetic data, ultimately achieving performance on the real test set that was better than a pure real-data baseline.
50%