2026-10-14T15:21:29Z https://www.scidb.cn/oai

10.57760/sciencedb.41685 2026-07-13T13:52:51Z

2026-07-13 Simulation Dataset for Deuterium Metabolic Imaging (Augmented with Brain Tissue Priors and Random Patterns) doi:10.57760/sciencedb.41685 en This dataset is generated using the bSSFP‑DMI forward signal model to provide simulated training data for metabolite separation under short-readout conditions. Metabolite spatial maps (HDO, glucose, lactate) are synthesized by linearly combining brain tissue segmentation masks (CSF/GM/WM) from the IXI dataset to mimic tissue-dependent distributions, while B₀ field maps are generated using a multi-scale spline random-field model to ensure spatial smoothness. To increase diversity, phantom samples with randomized geometric shapes and metabolite patterns are included, and complex circular Gaussian noise is added to the time-domain signals to simulate low-SNR acquisitions. Random T₁/T₂ relaxation perturbations are also applied to each sample.The data have a fixed map size of 32 × 32. The time-domain signal per voxel contains 128 sampling points (bandwidth 4065 Hz, dwell time 0.246 ms). Different readout lengths (M = 128, 72, 48, 36) can be obtained by simply taking the first M points from the 128-point signal.Each sample is stored as two mat files (.mat): the ground-truth file of size [Batch, 4, 32, 32] (channels: HDO, glucose, lactate, and B₀ map) and the signal file of size [Batch, 128, 32, 32] (128 time points per voxel). All data are complete with no missing entries, and signal values are within valid dynamic ranges. Deuterium metabolic imaging; Balanced steady-state free precession; Simulation dataset; Metabolite mapping; Deep learning; CycleGAN; Magnetic resonance spectroscopic imaging Bao Qingjia Huang Shuang RESTRICTED dataset http://www.doi.org/10.1016/j.mrl.2026.200272 Science Data Bank