Possibility to speed up the loading of nilRead?

[Laurens_de_Groo]

We're installing and configuring nilRead in Radboudumc to integrate with Epic, comparable to our current integrations. One of the worrying aspects noted in testing and validation is the loading time of studies when accessed from Epic. The total loading time of, say, 1000 slice CT scans is over 10 seconds which is unacceptable (most of our studies are way bigger than that and comparing solutions manage this within 1-2 seconds).

One of the parts seemingly causing this loading time, is the time it takes to start up nilRead in the browser. So our question would be: is there a possibility to speed this up by for example pre-loading certain aspects into memory?

[Jen_Tucker]

Hi Laurens,

There are a couple of options for you.  Tahir Naeem is going to be contacting you for further discussion.

Thanks,

Jen Tucker

Enterprise Imaging Product Manager

[Tahir_Naeem1]

Hi Laurens,

As discussed with Dorien Reijnders on project status call on 18th August 2021. Can you please provide feedback to the following queries:-

• Was the customer invocation undertaken on the test or production environment.
• Was IIS being used on the environment prior to capturing NIL start-up times? If not, first request to IIS will traditionally be slow. 

[Laurens_de_Groo]

Hi Tahir, thanks for replying, I was polling my inbox for your mail and disappointed to have missed it. 

The observed timings are from the integration in Epic TST --> nilRead TST. But we've seen the same behavior in PRD also.

We can reproduce this quite easy even after starting up multiple times after each other, the same delays are present. So it doesn't seem to be an IIS issue. Also, it's not about the first response from IIS, as we can see nilRead loading stuff on screen quite fast but then it keeps loading...

We're more than happy to show you the delays in a shared-screen meeting, if that helps.

[Jos_Slenter]

Hi Laurens,

Knowing your implementation, I can say that even with a direct integration into the VNA, the image building is slow. The rendering of a CT 1000+objects > over 10 secs, CT 5000+ objects > over 60 seconds, CT 15000+ objects > over 6 min. Typical the loading from archive to cache is over 1-2 ms per object. All the other time is needed to render in memory the study, which is nilread related. While rendering a study, the processor is normal reacting, memory consumption is normal. No thresholds etc

A second very evil behavior is the showing of the study. If you open a very large study, the image loading bar will show 100%. Nevertheless, not all series are shown. This never occurs with studies with a small object count.

So I'm curious what Hyland can say about rendering times and how to improve them.

@Tahir Naeem : If you have questions please feel free to sent a message....