Multi-modal brain MR image simulation data

The simulated Brain MR image database is based on the Bloch equations analytical solutions for Earnst angle solution, spin echo signal, and inversion recovery spin echo signal for brain MRI. A realistic multi-modal simulation framework is designed by incorporating patient-specific phantoms and Bloch equations-based analytical solutions for fast and accurate MR image simulations. The framework generates realistic-looking images with controllable imaging parameters such as SNR, TR, TE, and flip angle as well as MR tissue properties such as T1, T2, and proton density. A diverse brain image database with variable image appearances is simulated using the framework. To create our virtual population, we change three categories of parameters: i) anatomical parameters, ii) MR tissue properties and iii) imaging parameters. To create virtual subjects with natural variable anatomical representation, we utilized the self-derived patient specific phantoms (20-class brain classification labels acquired on real HCP T1w 3D brain MRI data). We vary T1 and T2 relaxation times and proton density for the 20 tissues visible in the field of view. We slightly alter repetition time (TR), echo time (TE) and flip angle for gradient echo, spin echo and inversion recovery sequence. Complex Gaussian noise is added in k-space to achieve variations in signal-to-noise-ratios. Tukey window filtering in the frequency domain (k-space) is applied to avoid ringing artifacts before FFT reconstruction. The images and labels are stored in NIFTI format. The first version of the multi-modal simulated database for Brain MR images, includes 200 patient specific virtual subjects with variable parameters. Each subject is provided with its corresponding ground truth label map including all simulated tissue types; Gray Matter, White Matter, cerebellum, corpus callosum, hippocampus, brainstem, pons, amygdala, fornix, putamen, thalamus, globus pallidus, caudate nucleus and accumbens, third ventricle, lateral ventricle, CSF and septum pellicudum. A detailed description of the tissue types and label information for simulated tissues is provided as a separate text file named label_info.txt.本模拟脑磁共振（Magnetic Resonance, MR）图像数据集基于布洛赫方程（Bloch equations）的解析解，涵盖用于脑磁共振成像的恩斯特角（Earnst angle）解、自旋回波信号以及反转恢复自旋回波信号的解析形式。本研究设计了一种贴合真实场景的多模态模拟框架，通过融合患者个体化体模与基于布洛赫方程的解析解，实现快速且精准的磁共振图像模拟。该框架可生成具有真实视觉效果的图像，且支持对成像参数（如信噪比（Signal-to-Noise Ratio, SNR）、重复时间（Repetition Time, TR）、回波时间（Echo Time, TE）与翻转角（flip angle））以及磁共振组织特性（如T1弛豫时间（T1 relaxation time）、T2弛豫时间（T2 relaxation time）与质子密度（proton density））进行灵活调控。基于该框架，我们模拟得到了具有多样化视觉表现的脑磁共振图像数据集。 为构建虚拟受试人群，我们调整了三类参数：① 解剖学参数；② 磁共振组织特性参数；③ 成像参数。为生成具有自然可变解剖结构特征的虚拟受试者，我们采用了自研的患者个体化体模——其标签为基于真实人类连接组计划（Human Connectome Project, HCP）的T1加权三维脑磁共振图像数据得到的20类脑分类标签。我们对视野内可见的20类组织的T1弛豫时间（T1 relaxation time）、T2弛豫时间（T2 relaxation time）与质子密度（proton density）进行调整。我们针对梯度回波、自旋回波与反转恢复序列，对重复时间（Repetition Time, TR）、回波时间（Echo Time, TE）与翻转角（flip angle）进行小幅调整。我们在k空间中添加复高斯噪声，以实现信噪比（Signal-to-Noise Ratio, SNR）的多样化调控。在快速傅里叶变换（Fast Fourier Transform, FFT）重建前，我们会在频域（k空间）中应用图基窗滤波（Tukey window filtering），以避免振铃伪影（ringing artifacts）的产生。图像与标签均以NIFTI格式存储。 本多模态脑磁共振模拟数据集的首个版本包含200例参数可变的患者个体化虚拟受试者。每例虚拟受试者均配有对应的真值标签图（ground truth label map），涵盖所有模拟组织类型：灰质（Gray Matter）、白质（White Matter）、小脑、胼胝体、海马体、脑干、脑桥、杏仁核、穹窿、壳核、丘脑、苍白球、尾状核、伏隔核、第三脑室、侧脑室、脑脊液（Cerebrospinal Fluid, CSF）以及透明隔。关于模拟组织的类型与标签信息的详细说明，将以名为label_info.txt的独立文本文件形式提供。