2024 MRSI Data Processing and Quantification Challenge Synthetic Dataset
收藏资源简介:
该数据集是由多所国际知名研究机构(包括慕尼黑工业大学、埃因霍温理工大学等)联合创建的2024年MRSI数据处理与量化挑战合成数据集,旨在为磁共振波谱成像方法提供可复现的受控基准测试平台。数据集包含32个受试者级别的子数据集(24个训练集和8个测试集),每个子数据集均包含受污染的FID-MRSI数据、解剖图像、B0场图、元数据以及已知真实值的代谢物和成分信号,共有384个时间采样点。创建过程基于人类连接组计划的高分辨率解剖图像和场图,通过前向模型模拟代谢物、大分子、残余水、脂质、基线和噪声等成分,并考虑空间编码和B0不均匀性效应。该数据集主要应用于MRSI处理方法、干扰信号去除算法和代谢物量化方法的开发与比较,解决了实验数据缺乏真实ground truth的难题,为算法评估提供可靠的基准。
This synthetic dataset for the 2024 MRSI Data Processing and Quantification Challenge was co-developed by multiple internationally leading research institutions, including the Technical University of Munich and Eindhoven University of Technology. It is designed to offer a reproducible, controlled benchmarking platform for magnetic resonance spectroscopic imaging (MRSI) methodologies. Comprising 32 subject-level subsets (24 training subsets and 8 test subsets), each subset contains contaminated FID-MRSI data with 384 temporal sampling points, anatomical images, B0 field maps, metadata, and metabolite and component signals with confirmed ground truth. The dataset is constructed using high-resolution anatomical images and field maps sourced from the Human Connectome Project (HCP). Forward models are utilized to simulate components including metabolites, macromolecules, residual water, lipids, baselines and noise, while accounting for spatial encoding and B0 inhomogeneity effects. This dataset is primarily applied to the development and comparison of MRSI processing methods, interference signal removal algorithms and metabolite quantification approaches. It resolves the challenge of lacking ground truth in experimental MRSI data, providing a reliable benchmark for algorithm evaluation.





