Clarifications Regarding Preprocessing and Variable Size Handling

I would greatly appreciate a response to these two issues:

a) Partial Abdominal Region in sCT Volumes: I found the stripped slices in several sCT volumes, particularly at the extremes. In these slices, the intensity distribution is severely affected, which impacts the preprocessing. According to the official preprocessing steps, face removal is performed (but only for hand and neck patients). Why are the regions in these abdominal slices incomplete? (The following figure shows some slices for two patients.)

b) Inconsistent Depth, Height, and Width: "Usually, the preprocessed datasets have consistent depth and spatial resolution. According to the official documentation, MR/CBCT, CT, and masks are cropped to reduce file size. Will the validation/test code require a consistent size, or will it be able to handle variable sizes?

Thank you in advance!

Hi Stak_GC,

Thanks for reaching out.

1) a clarification, we did not provide sCT. I assume you are referring to CT. Before providing an answer, could you give me a case number just to double check and then come back to you? What I can think of right now is that a mask has been applied and the mask is calculated as intersection of CT and MRI or CBCT. Since MRI and CBCT may be angulated compared to CT, what you see is the the effect you can get when applying the mask on registered/resampled images. If that occurs, you should observe such effects at the beginning and of the body contour, in the feet-head direction.

2) The challenges collects real-world cases, and the voxel size was made consistent, but the field-of-view may vary per case, and so also the matrix size (height, width and depth) of the volumes. Please note that you are allowed to further process the data as you wish if you believe this will give any advantage. Just note that training, val and test datasets are consistent si you need to apply the same preprocessing also during model inference.

Let me know what you think of this. We can look further into the preprocessing steps of a couple of cases for point 1) if you are not satisfied with the current explanation.

Warm Regards,

Matteo