Synthetic MRI pretraining for medical imaging classification.

Abstract

Deep learning (DL) models have reached remarkable achievements in medical imaging, but their performance heavily depends on the availability of large and diverse datasets. To overcome this limitation, transfer learning has emerged as a widely adopted solution, where models pretrained on large datasets are fine-tuned for specific medical tasks. Due to the scarcity of large-scale medical imaging datasets, most existing models are pretrained on natural image datasets such as ImageNet, while recent studies have explored highly complex models trained on large collections of medical unlabelled datasets. In this work, we generate and leverage a synthetic MRI dataset to pretrain DL architectures, demonstrating effective model learning with minimal computational cost. We evaluate our approach across multiple downstream tasks, including brain tumour classification and benchmark datasets from MedMNIST, including both 2D and 3D imaging modalities. Compared with ImageNet-pretrained, foundation, and self-supervised models, synthetic pretraining consistently improves feature representations and downstream performance. Overall, our approach outperforms competing methods, establishing a new state-of-the-art on the MedMNIST benchmark.