研究目的
To compare Fourier-based and SHORE-based compressed sensing reconstruction for accelerating diffusion spectrum imaging, and to evaluate different q-space undersampling schemes for robust data recovery.
研究成果
Fourier-based CS-DSI using the canonical basis as the sparsity transform outperforms SHORE-based CS-DSI in reconstructing diffusion signals and propagator-derived parameters, while both methods provide comparable orientational information. Isotropic q-space sampling is optimal for high acceleration factors. A modest CS acceleration factor of 2.3 is recommended for time-limited studies to ensure robust reconstruction with minimal artifacts.
研究不足
The study focuses on specific basis functions (Fourier and SHORE) and may not generalize to other bases. In vivo data were from a single participant with fixed sequence parameters, limiting generalizability to different resolutions or field strengths. The impact of image artifacts and noise on CS reconstruction was not fully investigated. Computational costs for high SHORE orders are high, and CSF regions showed higher errors in SHORE-based methods.
1：Experimental Design and Method Selection:
The study compares two compressed sensing (CS) reconstruction frameworks (Fourier-based and SHORE-based) and three q-space undersampling schemes (random Gaussian, uniform angular and random radial, isotropic) for diffusion spectrum imaging (DSI). CS theory is applied to recover full data from sparse measurements using l1-norm minimization.
2：Sample Selection and Data Sources:
Data were obtained from simulations (using Camino Monte-Carlo simulator for crossing fiber structures), diffusion phantoms (physical phantoms with 60° and 90° crossings), and in vivo human brain DSI (from a healthy female participant).
3：List of Experimental Equipment and Materials:
A 3T Siemens MAGNETOM Prisma MRI scanner with a 64-channel head-neck coil was used for phantom and in vivo data acquisition. Software tools included Camino for simulations, Dipy library for signal processing, and FSL for image correction.
4：Experimental Procedures and Operational Workflow:
Fully sampled DSI data were acquired or simulated, then retrospectively undersampled using the three schemes. CS reconstruction was performed using Fourier-based (with canonical basis or total variation) and SHORE-based (orders 6 and 8) methods. Reconstruction quality was assessed using normalized mean square error (NMSE), angular errors, and diffusion parameters (e.g., MSD, RTOP, NG).
5：Data Analysis Methods:
NMSE was computed between reconstructed and ground truth signals. Diffusion parameters and fiber orientation distribution functions (fODFs) were estimated using MAP MRI model and multi-tissue deconvolution. Statistical analysis included mean and standard deviation calculations across multiple instances for simulations.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
