ProstateX ids

Hi, Would it be possible to know which study ids (among the 1500 in the database) correspond to the 328 ProstateX studies ? Thanks in advance!

Hi Noëlie,

Due to the anonymisation procedure for the PI-CAI: Public Training and Development Dataset, we do not have a mapping between the anonymous ProstateX IDs and the anonymous PI-CAI IDs.

I'm curious, what would you need this mapping for?

Kind regards, Joeran

Hi Joeran,

Ok, thanks for your answer ! It is because some of ProstateX contours are already available (https://github.com/rcuocolo/PROSTATEx_masks/tree/master/Files/lesions/Masks). As long as we miss some contours for positive studies, I was wondering if we could increase the number of annotation thanks to this mapping ! It's ok, I'll try to compute the correspondance between PICAI and ProstateX :) Best regards, Noëlie

Hi Noëlie,

We can confirm that all ProstateX cases included in the Public Training and Development Dataset, were annotated by a human expert. So no advantage in leveraging lesion delineations from R. Cuocolo et al.. Also, during our annotation workflow, we spotted some errors and inconsistencies between the lesion delineations provided by R. Cuocolo et al., with respect to the original clinical findings reported for these ProstateX patients in our hospital system. Hence, we opted to annotate them from scratch and have provided the same through picai_labels.

Hope this helps.

It should be easy to map the IDs. Just calculate a crypto hash sum (e.g. MD5) of the 3D voxel data of each ProstateX T2 axial sequence and do the same for each PI-CAI T2 axial sequence. Finally, for each ProstateX case, look up the PI-CAI ID with the same crypto hash sum. Voila!

Well, that assumes PI-CAI organizers didn't change the voxel data.

Hi Nathan,

We have decided to provide the ProstateX → PI-CAI ID mapping. The mapping consists of the 328 ProstateX cases within the PI-CAI Public Training and Development dataset. You can find the mapping here: picai_labels/additional_resources/ProstateX-mapping.json.

The remaining cases from the ProstateX dataset were excluded during PI-CAI dataset curation, with exclusion criteria shown below.

ProstateX ID: Exclusion Reason ProstateX-0006_10-21-2011: Missing PI-RADS Information ProstateX-0047_11-17-2011: Missing PI-RADS Information ProstateX-0052_11-21-2011: Hip Prostheses ProstateX-0053_11-22-2011: N/A ProstateX-0067_12-06-2011: Missing Pathology Information in EPIC and PALGA ProstateX-0148_03-12-2012: Missing/Obscure Prostate ProstateX-0150_03-15-2012: N/A ProstateX-0151_03-18-2012: N/A ProstateX-0155_03-22-2012: N/A ProstateX-0182_05-17-2012: Missing PI-RADS Information ProstateX-0192_06-02-2012: Missing Pathology Information in EPIC and PALGA ProstateX-0204_02-22-2012: N/A ProstateX-0205_04-16-2012: N/A ProstateX-0216_02-16-2011: Missing PI-RADS Information ProstateX-0217_02-12-2011: Missing PI-RADS Information ProstateX-0218_02-18-2011: Missing PI-RADS Information ProstateX-0229_04-29-2011: N/A ProstateX-0315_03-24-2011: N/A ProstateX-0325_04-02-2011: N/A

Hope this helps, Joeran

Hi guys I'm a beginer in radiomic feature extraction by using CNN, can anyone to helpe me to accès to a methodologie to do that? Secondly, I want to acces to PI-CAI dataset (Image and anotations ) in nifti format

I usally use U-Net and other CNN methode for medical image segmentation.

Thank you all

Hi Souvenir,

We do not provide any tutorial for radiomics, but we did make our baseline models publicly available. These models might be useful to extract features. However, this is not a use-case that's supported by us.

The imaging dataset is made publicly available in MHA format, but you can convert these to nifti using tools like SimpleITK in Python. Also, you can follow our baseline solution tutorial, which converts the MHA Image Archive to the nnU-Net Raw Data Archive format, which consists of nifti images.

Hope this helps, Joeran