Clinical implementation

[Rajagopalan B]

Hi,
I am very much interested in decreasing human error in clinical practice and to increase the involvement of machine .so I am asking you to enlighten me what are simple and efficient ways to implement the machine (you can consider ML) in daily usage in clinical practice eg :- offline image checking , DVH parameters checking , daily QA last but not least analysing integrity of data collected

P.S:- This can be consider as an open discussion 😁

[Simon Biggs]

Hi Rajagopalan,

I've CCed in Mathew Cooper. He's built a machine learning tool that is designed to validate contours. It is still early stages, but he has successfully integrated it with a DICOM server and integrated it with our Monaco TPS, and it is successfully creating usable contours. It can be used to either receive DICOM CT files + RT Structure files and then provide feedback on how much agreement there is, or, if one so desires, it can be set up to receive CT DICOM files and it will then send those back with a newly created RT Structure file which contains contours created by the machine learning algorithm. It takes ~10 seconds on a GPU or ~1 minute on a CPU.

The work done by Matt was built on top of the work done by deep mind:
https://deepmind.com/research/publications/deep-learning-achieve-clinically-applicable-segmentation-head-and-neck-anatomy-radiotherapy

I believe right now, he is quite keen on having people provide him with clean anonymised datasets. PyMedPhys has a tool that can be used to anonymise DICOM files:

https://docs.pymedphys.com/ref/cli/dicom.html#anonymise

If someone did want to collate data for him to help him further improve his auto-contouring algorithm it would be a good idea to have a video call first to get a good understanding of what is required from you in the data preparation stages.

Cheers,
Simon

--
You received this message because you are subscribed to the Google Groups "PyMedPhys" group.
To unsubscribe from this group and stop receiving emails from it, send an email to pymedphys+...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/pymedphys/f6a8435f-24b8-4fbc-96c4-a5bdd7fbc382o%40googlegroups.com.

[Rajagopalan B]

Hi Simon ,

                  I have doubt whether I can use linear regression to analyse field size vs field output factor.kindly help me.with this topic .

               It's good idea to implement contour automation .And also I am two topics DVH parameter evaluation , offline image analysis and creating plan template .If I get stuck with python I will ask you the doubts. Thanks a lot for your help .

Rajagopalan B

[Simon Biggs]

Hi Rajagopalan,

Here is a paper I wrote that uses bivariate splines to create a fit for electron output factors:

http://dx.doi.org/10.1016/j.ejmp.2015.11.002

I personally have never attempted doing something similar for photon output factors.

The code from that paper is within pymedphys. You can see it below:

https://github.com/pymedphys/pymedphys/blob/c00fea77488e5919c53b475bcd217ba40ea1b5aa/pymedphys/_electronfactors/core.py#L23-L70

Also, this probably isn't the best forum for these sorts of questions.

Cheers,

Simon

To view this discussion on the web, visit https://groups.google.com/d/msgid/pymedphys/CABZ8T5kd9oJZRGVZ6cScEGiANOLX1dHEuoq7fnWMb1JdvX2oiw%40mail.gmail.com.