How to get json data?

[SE-Nickjackson]

Hi, thanks a lot for your work, I would like to ask how the json data is obtained, can you provide the corresponding python scripts?

[maxme1]

Hi! Do you already have an annotation in some format?

[shangboli]

你好作者，我有属于自己的脑出血nii数据集 我应该怎样得到json数据呢

[maxme1]

Hello! If you only have the images, you will need to annotate them somehow. We used an internal tool for that, but I think you can easily annotate the midlines with Slicer3D or MedSeg. I'm not sure whether MedSeg has a curve annotation tool, but Slicer3D does have one.

[shangboli]

谢谢作者 我还有个疑问 slicer曲线标注是 应该是2d还是3d的对于我们这个任务. 曲线注释完后应当保存问什么格式呢

[maxme1]

The brain midline is a surface, so I would annotate it as a sequence of 2D curves. Basically you should have a curve per axial slice, so that combined they create a surface. If you want to use the code from this repository, the data should be represented exactly this way - a sequence of 2D curves, one per axial slice. You can read more here on the data format.

[shangboli]

还有一个问题 作者 我现在找到如何注释曲线的方法了 但在3d图像上应该注释多少个切片的曲线 还有 nii图像输入尺寸有要求嘛

[maxme1]

You can use images of any size, there are no restrictions. Regarding the number of slices to annotate, it really depends on the curvature of the surface. You could annotate every 10th slice, and then linearly interpolate for the rest of the slices, for example. We used this library for interpolation.

[shangboli]

hi 我最近有了很多新的进展 很感谢您的耐心回答 但是我发现我标记的中线可能不是很准确 我看您的论文中 ct图像的对比度好像更好的可以看清中线 我可以知道您用的窗位窗宽值是多少吗

[maxme1]

Hello! We used MRI images in our paper, not CT, this is why the contrast is so much better. Unfortunately the window location and width are not translatable between MRI and CT. I would suggest trying several window parameters and then aggregating the annotations afterwards.