search

SeuTao / RSNA2019_Intracranial-Hemorrhage-Detection

Code for 1st Place Solution in Intracranial Hemorrhage Detection Challenge @ RSNA2019
413 stars 113 forks source link

Windows actually used in preprocessing #3

Closed meegoStar closed 4 years ago

meegoStar commented 4 years ago

Hi, thanks for sharing your great work. According to your README, brain, subdural, & bone windows are used when preprocessing dicoms to pngs.

window

But in prepare_data.py, it seems that window parameters are extracted from dicoms' metadata, no explicit windows given. And in dataset.py, s-1, s, s+1 images are simply loaded, concatenated, applied data augmentation, no other preprocessing introduced. So I just wonder are brain, subdural, & bone windows actually used?

SeuTao commented 4 years ago

Hi, this is our solution writeup. We have three pipelines for single 2D CNN training (3 different preprocessing methods). The uploaded code is one of the three which only applies the 2nd preprocessing method (Spatially adjacent 3 slices with one window). You can get similar performance with the other two preprocessing methods. Ensemble of the three pipelines can give you a boost on final score.

meegoStar commented 4 years ago

Got it. Thank you very much!!!