AI integration in Bahmni (call for ideas)

Software Bahmni
1

Hi,

It has been a while since I’ve been on this channel, please forgive my ignorance on progress made here.

Context

A hospital that uses Bahmni is running an experiment to see how artificial intelligence can help doctors with consultation. Here is a description of the experiment.

  1. Nurses collect detailed history from patients in the form of an audio file. This happens in Bengali/Hindi languages
  2. The audio is sent manually to a Google Gemini “Gem” that takes in this information and suggests next steps
  3. The output is stored and then compared to notes from a doctor who then does the consultation
  4. The experiment is also reversed with the doctor talking to the patient first, and the nurse talking next

Based on what I hear, preliminary findings show that AI can assist, and sometimes even outperform doctors in constrained settings.

Need

They are looking to see if it is possible for this mechanism to be tightly integrated with Bahmni. The workflow would look something like this (in my imagination).

  1. Bahmni “listens in and records” a conversation between a provider and patient
  2. Once the conversation is over, the audio, along with patient history available in Bahmni, is sent to an LLM.
  3. The LLM analyzes and provides next steps - investigations/drug orders etc.
  4. The LLM output is auto-populated into Bahmni (unsaved)
  5. The provider looks at the suggestions, makes any necessary changes required and saves on Bahmni

Benefits

  1. More comprehensive medical record with lower effort
  2. Suggestions from AI can potentially reduce errors during consultation

Questions

  1. Is there existing work that we can utilize for this purpose? I tried looking, but I could have missed something
  2. Would anyone else also be interested in a solution like this?
  3. What would it take to build a native experience for this workflow on Bahmni?
1 Like
2

Hi @vinay We are currently working on integrating diagnosis and drug simulation in Bahmni, so your idea of using audio-driven AI suggestions fits really well with the kind of intelligent assistance we’re exploring.

3

This would be a great addition to Bahmni @vinay

We integrated with Medispeak last year but that was just to fill the forms. Intellidoctor team from Brazil also integrated with Bahmni with an ambient AI solution (notes and demo video are here).

It would be a good idea to include this discussion in our PAT call agenda for tomorrow.

4

Thanks @akhilmalhotra and @santosh Would be glad to discuss in today’s PAT call. I will go through the demo video by then and also talk to folks doing the project to see if they would be interested in joining in.

1 Like
5

@vinay welcome back.

This is exactly the experience that modern clinical solutions should bring in - provide “freedom” and “time” to clinician, while ensuring clinical safety and propriety. We have had some discussions and debates on what “role” an open source project like Bahmni ought to play while embedding such capabilities.

Would love to discuss this in the PAT call.

Please bring up

1 Like
6

Hi Vinay,

This is interesting and would like to be part of this discussion.

Can we also try image processing tools to capture information from ID cards for registration, for medications entry and for notes analysis?

Gajanan