Submillimetric COVID-19 CT dataset and automatic lung tissue labeling extension

Hi all,
I apologize if I’m posting in the wrong section (if so please remove/move this post).

We just released a free (non commercial use) dataset made of 62 submillimetric CT scans of COVID-19 positive patients.
Automatic lung tissue labeling and human-based clinical score are provided as well.
Maybe in this difficult historical period, this dataset can be useful for your research.

Here the article:

My 2 cents.

Paolo

8 Likes

Dear Paolo,

That is great, thanks for your contribution.

Are there other clinical data such as survival outcomes, ICU admission or other (anonymized) data available as well?

Best,
Justin

Hi Paolo,

Thank you, this looks like an amazing source.

We recently developed a Lung CT Analyzer extension for 3D Slicer for COVID-19 assessments. It can be installed within Slicer from the extension manager.

The sources can be pull from here: GitHub - rbumm/SlicerLungCTAnalyzer: This is a 3D Slicer extension for segmentation and spatial reconstruction of infiltrated, collapsed, and emphysematous areas in chest CT examinations.

We will definitely check out your data, comments and ideas for the extension are always welcome !

Best regards

Rudolf Bumm, Chur, Switzerland

Thanks @Justin !
We are trying to retrive these data, but I can not promise anything.

@rbumm I saw your extension…that’s really great!
The aim of our paper was to release high quality images (on internet a lot of people published png-like images…) alongside with some clinical data and a way to automatically get some labels from CTs.
Let me know if I can provide additional information (as already said we are trying to retrive additional clinical information).

Best,
Paolo

1 Like

I would be very interested in how the community could help to enlarge the dataset.

@lassoan @jcfr do you think it would be interesting for the community to have an extension that, given a chest CT, automatically labels lungs tissues on basis of the intensity (healthy, ground-glass opacities, and consolidation)?
It can be used for pneumonia (COVID too).
The methodology is explained in the paper linked in the first post (it’s a GMM strategy basically).

We could call it “Lung density segmentation”.

Yes, sure, I expect that this would be useful. If you don’t want to create an entire new extension for this then it could be added to the LungCTAnalyzer or ChestImagingPlatform extensions.

This sounds interesting, the LungCTAnalyzer does that to some extent, and it would be great if you could add ideas or add functionality yourself.

1 Like

I submitted a pull request.

I added a python script

on the LungCTAnalyzer GitHub that allows analyzing all of your 62 datasets in a row in ~60 minutes (gaming laptop).

As a prerequisite, lung masks have to be prepared once for every CT.

My results of the initial CT’s of all 50 cases are summarized in this google spreadsheet:

Thresholds:
logic.setDefaultThresholds(-1050,-990,-650,-400,0,3000)

Without knowing any clinical details on the patients, LTCA predicts a severe clinical course (20-50% affected lung) in

D:/Patients5\45\CT.nrrd
D:/Patients5\46\CT.nrrd
D:/Patients5\7\CT.nrrd
D:/Patients\35\CT.nrrd
D:/Patients\33\CT.nrrd
D:/Patients\15\CT.nrrd
D:/Patients5\47\CT.nrrd
D:/Patients\40\CT.nrrd
D:/Patients\37\CT.nrrd
D:/Patients\3\CT.nrrd
D:/Patients\4\CT.nrrd
D:/Patients\12\CT.nrrd
D:/Patients\29\CT.nrrd

and a fatal one (> 50 % affected lung) in

D:/Patients\10\CT.nrrd
D:/Patients\14\CT.nrrd

Dear all,
today I updated our COVID-19 database.
Now it is made of 81 patients (in total 93 CTs since some patients have follow-up scan) and information about intensive care unit were added as well.

@rbumm @Justin

HTH,
Paolo.

3 Likes

Very interesting @PaoloZaffino - would it be possible to add mortality data too?
Thank you
Best regards
Rudolf

Not easy to get, but I’ll do my best.

Paolo

1 Like