SMRI: Next-generation MRI simulation platform for training data generation in the era of AI

Motivation: Current MRI simulation remains significantly limited by speed, restricting the use of synthetic training samples in deep learning-based MRI methods. Goal(s): To develop an ultra-fast, versatile, cross-platform, and user-friendly MRI simulation platform for training data generation. Approach: The SMRI platform was developed to support rapid MRI data generation by accelerating Bloch simulations. It enhances data diversity through deep learning-based MRI modality transformation and includes a user-friendly interface for ease of use. Results: The SMRI platform delivers a 10- to 100-fold improvement in simulation speed over existing software, enabling diverse downstream tasks such as motion correction and quantitative MRI. Impact: The ultra-fast, cross-platform, and user-friendly SMRI platform was developed for deep learning training sample generation, providing available and sufficient datasets for various deep learning-based MRI tasks within an acceptable time.