面向MRF图重建的FPGA超快速神经网络硬件加速
硬件体系结构
2025-06-30 v1 计算机视觉与模式识别
仪器与探测器
摘要
磁共振指纹 (MRF) 是一种快速定量MR成像技术，通过单次采集提供多参数图。神经网络 (NN) 加速重建，但需要大量资源进行训练。我们提出一种基于FPGA的NN，用于从MRF数据实现大脑参数的实时重建。NN的训练预计需要200秒，显著快于标准CPU-based训练，速度可快达250倍。该方法可能使大脑分析在移动设备上实现实时，从而革新临床决策和远程医疗。
引用
@article{arxiv.2506.22156,
title = {Hardware acceleration for ultra-fast Neural Network training on FPGA for MRF map reconstruction},
author = {Mattia Ricchi and Fabrizio Alfonsi and Camilla Marella and Marco Barbieri and Alessandra Retico and Leonardo Brizi and Alessandro Gabrielli and Claudia Testa},
journal= {arXiv preprint arXiv:2506.22156},
year = {2025}
}
备注
8 pages, 2 figures, to be published in conference proceedings of SDPS 2024: 2024 International Conference of the Society for Design and Process Science on Advances and Challenges of Applying AI/GenAI in Design and Process Science