Inquiry Regarding Correction of CBCT Images in Dataset 2BA078

Dear SynthRAD_2023 Community,

I am writing to express my concern about the CBCT images in validation dataset 2BAO78. The HU values within the masked region consistently appear as 3000, raising doubts about the dataset's accuracy. Therefore, I would like to inquire whether there are plans to rectify this issue or provide an updated version of the dataset with corrected CBCT images.

Best regards, Chuanpu Li

Dear Chuanpu Li,

Thank you for reporting this issue with the dataset. We will check this case and come back to you with an answer on how we are going to handle this issue as soon as possbile.

Kind regards, Adrian

Dear Chuanpu Li,

Thanks again for noticing the issue. We have corrected the problem, and made a new version of the validation dataset, you can find it at https://doi.org/10.5281/zenodo.7868168.

What happended on this case? Such case had a different orientation (only case in the whole dataset), that has been manually adjusted in itksnap. While doing so, the data type has also been modified leading to a shift in image voxel intensity in the pre-processed image.

Please, let us know whether the new version of the data solve the issue. You can re-upload yopur predictions, I am curious to see the effect (it will surely be better than the current output).

I hope this hick-up did not interrupt the fun while diving into SynthRAD2023.

Have a great continuation,

Matteo