Recent Updates of the M4Raw Dataset and Applications in Evaluating MRI Denoising Methods

Yi Li, Mengye Lyu, Guanxiong Luo, Jingwei Guan, Jingyu Li

Keywords: Low-field MRI, Image denoising, Data-driven methods

Abstract: This paper presents the new multi-channel k-space dataset M4Raw acquired using low-field MRI. The M4Raw dataset comprises brain data from 183 subjects, each with 18 axial slices and three contrasts: T1-weighted (T1w), T2-weighted (T2w), and fluid attenuated inversion recovery (FLAIR). Additionally, the paper provides a description of the recently released test subset, as well as various denoising methods applied to the M4Raw dataset, demonstrating its potential applications in image denoising. Multiple deep learning methods trained on the M4Raw dataset, including traditional denoising networks and those using transformer modules, have been employed, achieving high-quality denoising of low-field MRI images. The M4Raw dataset not only facilitates the development of data-driven methods for low-field MRI denoising but also serves as a benchmark dataset for comparing different methods.

Submission Number: 115