search

pydicom / deid

best effort anonymization for medical images using python
https://pydicom.github.io/deid/
MIT License
140 stars 43 forks source link

I'm interested in validation of ocr deid tool #172

Closed marcheschi closed 3 years ago

marcheschi commented 3 years ago

Hi I'm interested in validation of ocr deid tool. How is it possible to contribute? Thank you Paolo

vsoch commented 3 years ago

Did you mean "our" deid tool, or "OCR" optical character recognition? What tool are you talking about?

marcheschi commented 3 years ago

I 'm interested in the OCR cleaning method . I see in your github deid page that you are looking for collaborators ? Paolo

vsoch commented 3 years ago

This repository? It’s not really related to deid. https://github.com/pydicom/dicom-cleaner. The library itself needs more work and the validation so if you want to take the helm I’d be happy to support.

marcheschi commented 3 years ago

Maybe you misunderstood me I do not want the helm, I can make tests in order to validate the tool on real world US DICOM images for example. In my opinion, a tool like this to deidentify burnt image annotations is very useful. Paolo

vsoch commented 3 years ago

Yes! And I mean to say it would be great if you want to do that.

marcheschi commented 3 years ago

Ok then . How to start? Maybe it is better to switch to a gitter chat. Paolo

vsoch commented 3 years ago

You probably need a gold standard set that has different configurations and modalities, and to manually label it and then run the tool to see how well it did.

marcheschi commented 3 years ago

Yes this could be a valid approach, to have various configurations applied to different modalities and to different manufacturer. I think the first step for me is prepare a VM with all necessary tools to make tests. I installed for this purpose a Solaris vm with Python 3.8.3 I installed deid with pip install deid. Or I have to install a git version in order for you to make modification to the code and for me to apply them to the tests?