Generating Synthetic Magnetic Resonance Images for Deep Learning Applications with Limited Labeled Data

• Caio dos S. Felipe UFCSPA
• Dieine E. B. Schiavon UFCSPA
• Thatiane A. P. Alva UFCSPA
• Carla D. L. Becker UFCSPA

DOI: https://doi.org/10.5753/sbcas_estendido.2025.7825

Resumo

This study addresses medical imaging data scarcity in AI development by using Progressive Growing of GANs (PGGAN) and transfer learning (TL) to generate synthetic MRIs of frontotemporal dementia variants. Our pre-trained foundation model achieved high anatomical fidelity (FID: 21.6008), while TL significantly improved image quality compared to models trained from scratch, reducing FID scores from ~115 to ~40. Although clinical validation is still needed, this framework offers a promising approach to expand limited medical datasets while maintaining anatomical accuracy for rare disease analysis.

Referências

Vonder Haar, L., Elvira, T. and Ochoa, O. (2023) “An analysis of explainability methods for convolutional neural networks”, DOI: 10.1016/j.engappai.2022.105606.

Ehsani-Moghaddam, B., Martin, K. and Queenan, J.A. (2021) “Data quality in healthcare: A report of practical experience with the Canadian Primary Care Sentinel Surveillance Network data”, DOI: 10.1177/1833358319887743, January–May.

Ranjbar, A. and Ravn, J. (2023) “Data Quality in Healthcare for the Purpose of Artificial Intelligence: A Case Study on ECG Digitalization”, DOI: 10.3233/SHTI230534, June.

Celard, P. et al. (2023) “A survey on deep learning applied to medical images: from simple artificial neural networks to generative models”.

Karras, T., Aila, T., Laine, S. and Lehtinen, J. (2018) “Progressive Growing of GANs for Improved Quality, Stability, and Variation”, [link].

Grigoryev, T., Voynov, A. and Babenko, A. (2022) “When, Why, and Which Pretrained GANs Are Useful?”, [link].

Mo, S., Cho, M. and Shin, J. (2020) “Freeze discriminator: A simple baseline for fine-tuning gans”, [link].

Karras, T., Laine, S., Aittala, M., Hellsten, J., Lehtinen, J. and Aila, T. (2020) “Analyzing and improving the image quality of stylegan”.

Heusel, M., Ramsauer, H., Unterthiner, T., Nessler, B. and Hochreiter, S. (2017) “Gans trained by a two time-scale update rule converge to a local nash equilibrium”.

Borji, A. (2022) “Pros and cons of GAN evaluation measures: New developments”.

Frontotemporal Lobar Degeneration Neuroimaging Initiative (2010) “FTLDNI Database”.