Author: haoransun
Wechat: SHR—97
https://ida.loni.usc.edu/pages/access/search.jsp
1-实验数据集下载
T1W1-3D-MP RAGE:T1加权三维磁化强度预备梯度回波序列属于快速容积扫描技术,具有较高的空间分辨率和时间分辨率,信噪比高,伪影小,对 脑内结构(如白质、灰质和脑脊液)对比度良好,能三维显示人脑内部精细解剖结构,有利于显示小病灶及其细节,对神经系统疾病的诊断具有重要价值;同时也是获取正常人脑的三维可视化图谱的重要方法。
T1W1-3D-MP RAGE:T1W1 three dimensional magnetization prepared rapid acquisition gradient echo sequences.
挑选自己想要的数据进行下载,此处对AD/MCI/CN只选取年龄在60-85之间的人,每个人只选择一个3D图像
最后选出来3个数据集:
- My_Study_AD
- My_Study_CN
- My_Study_MCI
由于下载的My_Study_MCI男女数量差别过大,重新进行下载My_Study_MCI_Male和My_Study_MCI_Female,到本地进行数据合并。
自己挂梯子开始下载可能比较快一点,下载CSV,NIFTI,MetaData,一健下载
2-数据集性别/年龄分布
Dataset | Male | Female | Age_Round | Mean_Age | Count_Number |
---|---|---|---|---|---|
AD | 44 | 36 | 60 ~ 85 | 75.59 ± 5.86 | 80 |
MCI | 40 | 40 | 60 ~ 84 | 74.15 ± 7.12 | 80 |
NC | 52 | 48 | 70 ~ 85 | 75.83 ± 3.85 | 100 |
由于上面下载的数据头动矫正一直报错,网上说不同患者的头动参数差距过大,此处还是使用other 分享的。
后来知道是format是DCOM格式的,需要自己手动转换为nii格式,但是自己明明下载的就是nii格式。。。
ADNI1_Baseline_3T因为每个人可能超过两张MRI图像,所以用代码随机删除csv文中的条目。确保一个人只有一个MRI图像。
3-ADNI挑选有意义的数据进行下载
ADNI数据集本身提供处理过的图片
有一篇论文“Non-White Matter Tissue Extraction and Deep Convolutional Neural Network for Alzheimer’s Disease Detection”中涉及到的预处理的数据如下:
“We use the FDG-PET and MRI data downloaded from ADNI1 dataset with each pair of FDG-PET and MRI for same subject and captured at the same time. The MRI and PET images have undergone several preprocessed steps of research groups belonged to the ADNI.”
In detail, the MRI images are pre-processed by steps: gradwarp, B1 non-uniformity and N3. Gradwarp means correction of image geometry distortion due to gradient model, B1 non-uniformity is a correction procedure that uses B1 calibration scans to correct image intensity nonuniformity. Finally, a N3 histogram peak sharpening algorithm was applied to reduce intensity non-uniformity of images.
其中,对MRI图像进行预处理的步骤有:梯度偏差、B1非均匀性和N3。梯度偏差是指梯度模型引起的图像几何畸变的校正,B1非均匀性是利用B1校正扫描对图像强度非均匀性进行校正的一种校正方法。最后,采用N3直方图峰值锐化算法来降低图像的强度不均匀性
For the FDG-PET images, a procedure involving dynamic co-registering frames and acquiring averaging from baseline PET scan was conducted. PET images were reoriented as AC-PC correction into a standard 160x160x96 voxel image grid, having 1.5mm cubic voxels. These images underwent continued filtering with a scanner-specific filter function to procedure images of a uniform isotropic resolution of 8mm Full Width at Half Maximum (FWHM).
对FDG-PET图像,采用动态共配帧和基线PET扫描平均的方法。将PET图像重新定向为AC-PC校正,得到标准的160x160x96体素图像网格,体素为1.5mm立方。这些图像在半最大值(FWHM)条件下,通过一个特定于扫描器的滤波功能,对均匀各向同性分辨率为8mm的全宽图像进行连续滤波。
那么如何针对论文中提到的数据进行搜索呢?
我们首先搜索AD类的数据,且同时搜索MRI和PET模态的数据,请见上图。单击搜索(SEARCH)后如下图所示。
可以发现,受试者002_S_0619的图片都是MRI模态的,没有PET模态的,所以放弃这是受试者,通过滚动条,继续查找下一个受试者,因为我们的目的是寻找同一年内既有MRI图像又有PET图像的受试者,最终把这两种模态的图片进行配准等多模态分析,可以详见论文。接下来继续:
这里84.6和84.7岁视为85岁,且存在满足条件的图像,所以,就勾选上,然后单击Add to Collection按钮