Associated Data for “Self-Paced Learning-Driven Training-Free Deep Generative Magnetic Resonance Image Denoising”

This dataset contains the associated data for the article “Self-Paced Learning-Driven Training-Free Deep Generative Magnetic Resonance Image Denoising.” It supports the quantitative comparisons, parameter sensitivity analyses, weighting-strategy evaluation, and computational-efficiency analysis reported in the article. The proposed SP-DIP method divides noisy magnetic resonance (MR) images into overlapping patches and progressively incorporates patches into network optimization according to their fitting losses.

The denoised patches are subsequently returned to their original spatial positions and averaged pixelwise to reconstruct the complete image.The experiments mainly used brain and knee MR images. Data #1 comprises T2-weighted brain MR data with an original matrix size of 256 × 232 × 208, acquired using a 3.0 T MR scanner and a 12-channel coil. Data #2 comprises T1-weighted knee MR data with an original matrix size of 320 × 320 × 227, acquired using a 3.0 T Siemens MR scanner and a 15-channel coil.