med-air / DiffusionMLS

[IPMI'23] Diffusion Model based Semi-supervised Learning on Brain Hemorrhage Images for Efficient Midline Shift Quantification
MIT License
11 stars 1 forks source link

数据集问题 #2

Open shangboli opened 1 month ago

shangboli commented 1 month ago

你好作者,我有自己的脑出血数据集,应当怎样标注达到该论文的应用效果呢

peterant330 commented 1 month ago

Hi, The labeling protocol of the paper entails the marking of three points, the anterior falx, the posterior falx, and the point with largest deformation, usually one among the falx, the third ventricle, and the septum pellucidum. Fig. 1 from the paper shows an illustration. You should obtain the coordinate of these landmarks.

For each volume, you could choose multiple slices with significant deformation to label.

To further improve the accuracy, you can draw the whole midline, i.e., draw a curve to connect the anterior falx, the landmark with largest deformation, and the posterior falx, and replace the point supervision with a denser supervision by sampling multiple points from the curve.

shangboli commented 1 month ago

谢谢作者,我之前用3d slicer标注数据 您用的是哪个软件标注获取这些形变监测点的坐标的呢

peterant330 commented 1 month ago

谢谢作者,我之前用3d slicer标注数据 您用的是哪个软件标注获取这些形变监测点的坐标的呢

we use ITK-SNAP to label