Do Synthetic Brain MRIs Reliably Improve Tumour Classification? A StyleGAN2-ADA Class-Plane Augmentation Study on BRISC 2025
A recent study investigates the effectiveness of synthetic brain MRIs in improving tumor classification using StyleGAN2-ADA. The research found that while some classifiers benefited from augmented data, others did not show significant improvements. Overall, the utility of augmentation was dependent on the architecture and the ratio of real to synthetic images used.
- ▪The study tested twelve class-plane StyleGAN2-ADA generators on constrained BRISC 2025 partitions.
- ▪MobileViTV2 showed the clearest benefit, with a 1.02% absolute improvement in tumor classification accuracy.
- ▪The random forest classifier did not benefit from the synthetic MRIs, while the CNN showed inconsistent gains.
Opening excerpt (first ~120 words) tap to expand
Electrical Engineering and Systems Science > Image and Video Processing arXiv:2605.23094 (eess) [Submitted on 21 May 2026] Title:Do Synthetic Brain MRIs Reliably Improve Tumour Classification? A StyleGAN2-ADA Class-Plane Augmentation Study on BRISC 2025 Authors:José Rafael Noriega Cedeño View a PDF of the paper titled Do Synthetic Brain MRIs Reliably Improve Tumour Classification? A StyleGAN2-ADA Class-Plane Augmentation Study on BRISC 2025, by Jos\'e Rafael Noriega Cede\~no View PDF HTML (experimental) Abstract:Generative augmentation is often proposed as a remedy for small medical-image datasets, but synthetic images are only useful when they improve downstream task performance.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
