I have worked with PACS software for 8 years.

And I can tell for sure that AI is one of the most powerfull tools on a DICOM viewer.

I can give you an example that I have developed with two of my fellow coworkers, we trained an AI to recognize all the vertebrae in an sagital PET CT or X Ray and almost instantly show MD where was the region of interest, in fact we could diagnose it, if it wasn't against the law. Recognize Cobb angle and everything.

That's just one of many uses in a DICOM viewer,

to Raphael Luiz Koszalka

Sounds very interesting. Could you share it with me, also?

Michael Sadovsky , you mean the research?

Or share has another concept here? I am new at this forum.

As for the research unfortunately I can't, it's a proprietary software, you can check it out some videos: here: https://www.youtube.com/watch?v=bYLLSHKYoGY

But if you are interested in modeling dicom images in general I would recommend for you to take a look at OHIF: https://ohif.org/

And off course, if you feel like exchange some ideas it would be my pleasure.

I'll write a full text on my profile about it. I'll just need a few days. I gotta a few heavy work days ahead.

to Raphael Luiz Koszalka:

Research, or course. No problem in timing: fill free to meet your schedule, especially keeping in mind my post-covid status. I mean, I am absolutely relaxed and lazy at the moment... :):) Simply speaking, I have very few possibilities to work... :)

Sure Michael Sadovsky , I just need to recapture somethings. By heavy work I meant my actual job, doing another line of research in a brazilian bank.

But for sure I'll do it, it will be my pleasure.

Brazilian bank... Pretty well acquainted situation! :):) Hope, the bank has nothing to do with AI! :)

Michael Sadovsky hahaha oh right I am doing quantitative analysis from a finite set of clients for credit rating purposes.

I am using Gaussian Kringing Process to emulate a "linear" equation (smooth-linear) trying to avoid neural network bias, by emulating an infinite set.

My first option were just use evaluation function, but got some algorithm bias.

And there was also Markov as an option, but chose Kringing because I found out that this would give me best 'linear' unbiased predictions.

So, lot's of AI.

Sorry if my answer did not meet your standards, since I wasn't trying to write a paper just objectively answering your question.

2 Raphael Luiz Koszalka you might want to seek for the "entities" with significant deviation of real entity frequency from that one determined through Markov process. They seem to be the most valuable in informativity in behaviour.

Thanks for the advise, I am going to try this approach.

Look at this:

https://www.worldscientific.com/doi/10.1023/A%3A1014230928565

https://pubmed.ncbi.nlm.nih.gov/16802083/

could be handy.