Possible modality mislabeling in metadata (non-contrast scans appear enhanced)

[LWHYC]

Hi, thank you very much for sharing the PanTS dataset — it’s an excellent resource for pancreatic tumor research.

While examining the metadata, I noticed that the modality labels for some scans marked as non-contrast may not be accurate. For example, the cases 0000006 and 0000007 appear to show contrast-enhanced phases upon visualization.

So far, I’ve only checked the first few examples labeled as “non-contrast,” but they all seem to correspond to contrast-enhanced images. Could you please confirm whether this might be a metadata labeling issue?

Thank you again for your great work and for making this dataset publicly available.

[WenxuanChelsea]

Hi @LWHYC

Thank you for contributing to PanTS and for taking the time to point this out.

You are correct — cases 0000006 and 0000007 do appear to show contrast-enhanced phases. This is a metadata mislabeling issue. We will correct this issue in the next update.

Thank you again for helping us improve PanTS! Let me know if you notice anything else.

Thanks,
Wenxuan

[dlotfi]

Hi @WenxuanChelsea, thank you for the clarification and for addressing the metadata issue.

I wanted to add that I also ran a contrast phase prediction over the PanTS dataset using the TotalSegmentator contrast phase prediction tool.

This tool segments multiple organs and vessels and uses their median HU values in an XGBoost model to predict the post-injection time (pi_time), which is then mapped to a contrast phase (native / arterial_early / arterial_late / portal_venous).

Based on these predictions, it appears that approximately ~25% of the CTs have a predicted contrast phase that conflicts with the phase stated in the metadata.

The phase prediction results are attached for reference:
Te_mismatch_phases.csv
Tr_mismatch_phases.csv

Thanks again for maintaining and improving this dataset, happy to share any additional details if helpful.

[WenxuanChelsea]

Hi @dlotfi

Thank you for the detailed analysis and for sharing the contrast phase prediction results — this is very helpful.

We have been aware of this issue since it was first reported by @LWHYC and, since October, have asked radiologists to re-check the entire dataset to ensure accurate contrast phase metadata. This is a very time-consuming process, but it is necessary to ensure data quality. We aim to release an updated version before the Chinese New Year.

Thank you for your patience and for the significant time and care you’ve put into improving PanTS.

Best,
Wenxuan

[LWHYC]

Hi @WenxuanChelsea ,

I hope you’re doing well.

I wanted to follow up on the contrast phase metadata issue. In your previous reply, you mentioned that the radiologists were re-checking the entire dataset and that an updated version was planned.

Has the corrected version of the dataset been released, or is there an updated timeline for the metadata correction? We are interested in using the latest version and would greatly appreciate any update.

Thank you again for all the effort your team has put into maintaining and improving PanTS.

Best,
Wenhui