rcuocolo / prostatex_masks Goto Github PK

Hello, I'm sorry to bother you,I have downloaded the mask label of the lesion in the ProStatex dataset you uploaded a few days ago. Thank you very much for your work.
But I have a question: in PROSTATEx_masks - master/Files/lesions/Masks/T2 / ProstateX - 0000 - Finding1 - t2_tse_tra_ROI. Nii. Gz file Prostatex-0000 patients have only one lesion, and the label in the data set is on the slice numbered 9. I see the mask you made on the label 8, 9, 10 and 11 all have lesions, but 11 is obviously larger than 10 and the position is further to the right
I would be very grateful if you could give me your reply. I wish you a successful scientific research

For example, for the patient ProstateX-0003, based on the data in the image_list.csv we should regard the following directory : /ProstateX-0003/10-17-2011-MC prostaat kliniek detectie-mc MCPROSKL30-03010/3-t2tsetra-30967/

We have in total 21 DICOM files there.
On the other hand the masks have only 19 slices.

I would expect the same number of slices.

There are several other patients having such differences (They are small but it looks important for me.).

Thank you very much for your help.

Hi, very nice work. I found there are some unmatched size between 'mask_prostate', 'mask_pz', 'mask_tz' and raw T2 dicoms from ProstateX challenge. Would you please check? Part of the files is listed as following:
error size: ProstateX-0102 pz
error size: ProstateX-0102 tz
error size: ProstateX-0141 pz
error size: ProstateX-0141 tz
error size: ProstateX-0160 pz
error size: ProstateX-0160 tz
error size: ProstateX-0165 pz
error size: ProstateX-0165 tz
error size: ProstateX-0166 pz
error size: ProstateX-0166 tz
error size: ProstateX-0168 pz
error size: ProstateX-0168 tz
error size: ProstateX-0173 pz
error size: ProstateX-0173 tz
error size: ProstateX-0181 pz
error size: ProstateX-0181 tz
error size: ProstateX-0191 pz
error size: ProstateX-0191 tz
error size: ProstateX-0197 pz
error size: ProstateX-0197 tz
error size: ProstateX-0199 pz
error size: ProstateX-0199 tz
error size: ProstateX-0201 pz
error size: ProstateX-0201 tz
error size: ProstateX-0202 pz
error size: ProstateX-0202 tz
error size: ProstateX-0203 pz
error size: ProstateX-0203 tz

Hi,
we are working working with your label set and I have come across some errors in the zonal masks. Here is a list of whole prostate masks with more than one connected component:

ProstateX-0000.nii.gz has 3 ccs
ProstateX-0014.nii.gz has 2 ccs
ProstateX-0027.nii.gz has 2 ccs
ProstateX-0064.nii.gz has 25 ccs
ProstateX-0112.nii.gz has 2 ccs
ProstateX-0126.nii.gz has 2 ccs
ProstateX-0141.nii.gz has 2 ccs
ProstateX-0142.nii.gz has 2 ccs
ProstateX-0168.nii.gz has 5 ccs
ProstateX-0170.nii.gz has 2 ccs
ProstateX-0172.nii.gz has 2 ccs
ProstateX-0174.nii.gz has 2 ccs
ProstateX-0179.nii.gz has 2 ccs
ProstateX-0181.nii.gz has 2 ccs
ProstateX-0182.nii.gz has 4 ccs
ProstateX-0183.nii.gz has 5 ccs
ProstateX-0190.nii.gz has 3 ccs
ProstateX-0194.nii.gz has 2 ccs
ProstateX-0198.nii.gz has 4 ccs
ProstateX-0203.nii.gz has 3 ccs
ProstateX-087.nii.gz has 8 ccs

Most of these have similar errors in the seperate peripheral and non-peripheral zone masks. I have merged the peripheral and non-peripheral masks for each patient and here are the cases where these errors persist:

ProstateX-0000.nii.gz has 3 ccs
ProstateX-0014.nii.gz has 2 ccs
ProstateX-0027.nii.gz has 2 ccs
ProstateX-0064.nii.gz has 25 ccs
ProstateX-0087.nii.gz has 8 ccs
ProstateX-0112.nii.gz has 2 ccs
ProstateX-0168.nii.gz has 2 ccs
ProstateX-0170.nii.gz has 2 ccs
ProstateX-0172.nii.gz has 2 ccs
ProstateX-0179.nii.gz has 2 ccs
ProstateX-0183.nii.gz has 3 ccs

The errors range from small, single voxel abrations close to the actual mask to larger errors that are very far from the mask and have a large impact on distance-based metrics.

Thank you for an otherwise terrific dataset,
Jakob

Hi,

Is there any update on your publication to cite this amazing repo?

Thanks :)

Hi Renato,

I have counted the number of components in each of the lesion mask nifti files. Most have one component as intended. However some have additional errant elements, so would request they be refined:

ProstateX0020-Finding1-t2_tse_tra_ROI.nii.gz | 2
ProstateX0025-Finding1-t2_tse_tra_ROI.nii.gz | 4
ProstateX0029-Finding1-t2_tse_tra_ROI.nii.gz | 3
ProstateX0032-Finding1-t2_tse_tra_ROI.nii.gz | 3
ProstateX0033-Finding3-t2_tse_tra_ROI.nii.gz | 2
ProstateX0050-Finding1-t2_tse_tra_ROI.nii.gz | 2
ProstateX0063-Finding1-t2_tse_tra_ROI.nii.gz | 2
ProstateX0067-Finding3-t2_tse_tra_ROI.nii.gz | 2
ProstateX0085-Finding3-t2_tse_tra_ROI.nii.gz | 2
ProstateX0086-Finding2-t2_tse_tra_ROI.nii.gz | 2
ProstateX0095-Finding1-t2_tse_tra_ROI.nii.gz | 2
ProstateX0099-Finding1-t2_tse_tra_ROI.nii.gz | 5
ProstateX0099-Finding2-t2_tse_tra_ROI.nii.gz | 2
ProstateX0104-Finding1-t2_tse_tra_ROI.nii.gz | 2
ProstateX0106-Finding2-t2_tse_tra_ROI.nii.gz | 2
ProstateX0165-Finding1-t2_tse_tra0_ROI.nii.gz | 3

Another issue is that the two clinically significant ProstateX0199 lesion masks drawn overlap when combined into a single volume. Can the ProstateX0199 lesion masks be adjusted so that they do not overlap.

As always many thanks for your efforts,
Pritesh

Note that TCIA, which holds the PROSTATEx collection, has a process that allows users to submit analysis results, see https://www.cancerimagingarchive.net/analysis-results/. This way it would be easier to discover your dataset, you would get a DOI that can be cited, and masks will be easy to access and visualize along with the original images. Would be great if you could consider submitting this dataset to TCIA.

Hi, thank you for this really nice repository with masks for the ProstateX dataset! :D While I was preparing the data I noticed two small things inside the image_list.csv:

• ProstateX-0025: the ADC sequence number ends with 7a which should probably be 7?
• ProstateX-0113: has a sequence number of 9 while the ADC file in the ProstateX dataset has the sequence number 10. Was this just a typo?

Thanks for your help!

Hi Renato,

The prostate/zonal mask for PROSTATEx-0102 uses T2WI sequence 9 for reference, while the lesion mask uses T2WI sequence 5 for reference. Sequences 5 and 9 have different shapes (and therefore the masks drawn using them also), which is causing me some problems, since I use both zonal and lesion masks simultaneously for training my system. Could the prostate/zonal masks be drawn using the same sequence used when drawing the lesion masks? My checks tell me this issue exists for the following cases:

102, 141, 147, 153, 160, 165, 166, 168, 173, 181.

There may be more cases where different T2 sequences were used for drawing prostate/zonal masks and lesion masks, but the ones highlighted above are the specific cases where differences in shape are causing issue.

Hi!
Regarding the ADC mask for ProstateX-0142:
Finding3 (ProstateX-0142-Finding3-ep2d_diff_tra_DYNDIST_ADC0_ROI.nii) geometry does not correspond to Finding1, Finding2 or the DICOM file:

• Finding1 and 2 geometry : (84, 128, 19)
• Finding3 geometry: (384, 384, 21)

Also, the ADC masks array is "flipped" corresponding to the DICOM array for every patients -->
Mask: (84, 128, 19), DICOM: (128, 84, 19)
I'm applying transpose to the ADC mask array, so both array geometry corresponds to each other.

I'm experience that the ADC mask slice order is inverted relative to the ADC dicom files.

Let us take ProstateX-0151-Finding2 for example:
ADC MASK: ProstateX-0151-Finding2-ep2d_diff_tra_DYNDIST_ADC0_ROI.nii
ADC DICOM: 7.000000-ep2ddifftraDYNDISTADC-75959/1-01.dcm --> 1-23.dcm

Both has 23 slices. From my experience it looks like ADC MASK lesion in slice 8 does not correspond to the slice 8 in ADC DICOM, but rather it correspond to the slice 15 (23-8=15). And that is for every ADC mask lesion examples. So the order is flipped, or inverted.

Can you confirm this?

Hi! Thank you very much for making the contours publicly available. Could you release more details about who produced the lesion masks and the quality checking process. In particular, were the lesion masks produced by clinicians? If yes, how many clinicians and how many years of experience did they have in prostate cancer mpMRI. These details will be needed for clinical journals, for work submitted that used the masks.

Hi! I have one question about the dataset composition.
As I read in the description, this dataset comprises prostate MRI exams with PI-RADS score = 2+ lesions, I was wondering if it contains PI-RADS score = 2 lesions. Thanks~

Dear @rcuocolo ,

Thanks for sharing the great work.

Would it be possible for you to share a download link or the image nii files (that you used to annotate the lesion masks)?

Best regards,
Jun

Dear @rcuocolo
First of all thank you for your wonderful and great work.
at the beginning i did not understand what are doing via this work but then i found it like a treasure. I am new in medical field and i was confused between the medical imaging extension and also for their true mask segmentation which we need them to train any deep learning model. specially for the prostate X dataset it's little bit weird for me because i could not understand the files which it contains. now i can say that your perfect give me the prostate masks which i need them to radiomic extraction, for that i am contacting you because i have some request:

1. I want to extract the radiomic features from T2 MRI, so i need to use the original slices and it segmentation label, which mask i need to use the prostate or the lesion? and it's possible to convert the masks to nrrrd file?
2. then I want to classify the gleason grade group, can you advice me how i can use the masks to do that?
3. otherwise, i am sorry to ask you alot, can you explain me more about prostatex dataset and how i can use your work for radiamic features and then use the radiomics features to classify the gleason?

looking forward to hearing from you
Kind regards,

Hi, thank you for providing this perferct repository. Now, I want to conduct the Multi-parametric (ADC, T2) lesion segmentation task. And I have faced two main problems, hope you can help me.

1. Since the resolution of ADC and T2 is not the same, do you have any registration method to make position consistent of different modalities.
2. The annotated mask of different modalities (ADC, T2) is not in the same slice, if we assume that I have done the registration of those two modalities, which annotated mask of the modality should be the final ground truth for the leison segmentation? ( ADC? T2? By voting or average those two masks ?)

Hope you can help me.

Hi, can you provide the image list of the training and test sets ?
In the paper(https://pubmed.ncbi.nlm.nih.gov/33634932/), it was described that "In detail, 99 cases from the PROSTATEx dataset were used as the training set in a 5-fold cross-validation fashion and the remaining 105 patients as the test set. The latter includes patients who have undergone biopsy for suspect of csPCa." However, I checked the file "PROSTATEx_Classes.csv" and found there were 99 cases with biopsy information, while another 101 cases without biopsy information. It would be appreciated if you could provide a list of test sets.

Hi again Renato,

Were PI-RADSv2 scores or otherwise (blinded to Gleason score) assigned to the lesions during the contouring exercise? If yes, would be very helpful to have them released.

Many thanks,
Pritesh