Synthetic MRI pretraining for medical imaging classification.
This work generates and leverage a synthetic MRI dataset to pretrain DL architectures, demonstrating effective model learning with minimal computational cost and establishing a new state-of-the-art on the MedMNIST benchmark.