Intelligent Chart Summarization

@andremillet,

That web link I gave you is limited. The goal is to get the docker image working and run it on AWS so we can use a web page for now.

Things are looking good. My fixes were checked into the docker repo and the author is helping me get the full thing working… very complex problem to solve but it’s coming!

  • m

Note I am at work so I will get back to you with a proper reply, but I was so excited to hear about your updates that I am writing back quickly!

1 Like

no worries!
if you need a faster response, send me a whatsapp msg
I am always on the move as well :sweat_smile:

cTAKES docker is now done!!! I will put together a video on how you can test it @andremillet

Please give me some time to do that.

1 Like
1 Like

very cool video. this is also a really nice real life docker tutorial.
-brady

Glad it was helpful!

could the installation run on AWS as well?
i mean, could i test it via web?

Just an update for anyone following this project. @andremillet is setting up the software on his machine and we discussed the need for running the pipeline via AWS. The idea is that one would submit clinical text to an endpoint and the output would be either the iml file or text printed to the screen via org.apache.ctakes.core.cc.pretty.plaintext.PrettyTextWriterUima.

I will hold on doing this work at the moment because we need to test out the software to see if we are happy with the results.

One potential block is that I don’t think the pipeline is using ICD10 by default. I’ll have to look into how to get it to produce ICD10 along with the SNOMED, RXNORM, etc.

UPDATES:

  • @toolbox has joined the project
  • @andremillet is getting the Docker containers running on his machine
  • @toolbox is getting the Docker containers running on his machine
  • @toolbox has suggested that we use cTAKES for smart “medical autocomplete” when writing notes. For instance, when typing “tylenol”, the related code shows up in a tooltip.
  • @toolbox and I are looking into parsing XMI output
  • I am looking into the programatic API (hasn’t been tested)

-m

1 Like

in the ‘smart medical autocomplete’ we can imagine two things:
1- the auto complete google uses os search box
2- a reference to a database offering things like the active formula of the
reference medicine (typing Tylenol - ‘would you like to write paracetamol
instead’)
in Brazil we are not even encouraged to use comercial names. I believe it
happens in all south america

Was kind of tired when recording this, but hopefully it makes sense! Keep in mind that my tangent at the end has more to do with the fact that this parser will most likely be ran in a Docker container and getting data from a container volume is something I’m not experienced with.

2 Likes

We have a great team: @andremillet, @toolbox, @JBW, and me!

Here’s some early work on the SNOMED and RXNORM parsing for anyone following along: https://gist.github.com/MatthewVita/5de7971e1adfb8724bdb989aa23317de

1 Like

@JBW are you still interested in working on this or are you deep in the analytics project? Just wanted to check before listing you as a teammate

Hi @andremillet, @toolbox, and @brady.miller,

One of the most difficult parts of this project is figuring out how to display the information in a useful way. Here is a VERY early mockup where a user can expand the codes after they were processed from a dictation encounter form:

Please provide much feedback!!!

-m

cc: @robert.down, who may have input from the clinical side :slight_smile:

I thought of some better UI stuff for this. Going to make use of color coding and hover overs… will report back in the coming days with more screens :slight_smile:

I just finished up my first version of the cTAKES ontological concept mention parser. It is a python program found here: https://github.com/MatthewVita/cTAKES-Concept-Mention-Parser - I still need to hook it up to the UIMA pipeline as a subscriber, but it works with just one-off XML (from cTAKES) files for now.

Please follow along to see what the parser can do:

  1. View the patient text here: https://github.com/MatthewVita/cTAKES-Concept-Mention-Parser#example
  2. View the outputted XML from cTAKES: https://github.com/MatthewVita/cTAKES-Concept-Mention-Parser/blob/master/samples/data.xml
  3. View the outputted JSON (structured and well formed): https://github.com/MatthewVita/cTAKES-Concept-Mention-Parser/blob/master/samples/data.json

I think you’ll find working with this new data structure will be much easier than the XML. While the program is not perfect (much more to add), it is a good “version 1” program.

Please let me know if you have any questions.

Working on the ICD10 dictionary… I got help from the core maintainers

Look at these realistic files… the JSON output is so much lighter than the crazy XML: