feat: new healthcare use case unlocked with genAI

[elhadik]

Medical coding errors are common and can lead to denied claims, delayed payments, and even patient safety issues. GenAI can help address these issues by suggesting accurate medical codes, automating tasks, and improving standardization. With a human-in-the-loop approach, this can lead to faster claims processing, better reimbursement, and reduced coder burnout. Google's large language models (LLMs) have the potential to significantly automate the medical coding process, improving both efficiency and accuracy. By first preprocessing medical records using healthcare APIs to identify and extract relevant terms, patterns, and relationships, we can then leverage LLMs to suggest appropriate codes for diagnoses and procedures based on physician notes and reports. This can not only save coders time and effort but also reduce errors and inconsistencies, leading to improved reimbursement rates and better financial management for healthcare providers.

**It is always recommended to have a human-in-the-loop when dealing with medical applications. This colab is meant to accelerate the medical coding processing and not to fully replace a human coder. ***

Read the below section to learn more about different kinds of medical codes:

[google-cla[bot]]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[review-notebook-app[bot]]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[holtskinner]

@elhadik Thanks for the PR, looks good overall! I just have some comments/suggestions to make it easier to read/more optimized.

Note - It looks like your GitHub account does not have the correct email attached to it, which makes the CLA check fail. Can you update your account to have the email ***@google.com instead of ***@googlecom?

[holtskinner]

Can you change the name of the Notebook to be all lowercase and more concise? Maybe react_gemini_healthcare_api.ipynb ?

And be sure to change the in-notebook links for Open in Colab, etc.

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

holtskinner commented on 2024-01-17T14:17:57Z ----------------------------------------------------------------

This markdown table formatting is off in

the review tool. Can you change this to

use bullet-points?

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

holtskinner commented on 2024-01-17T14:17:58Z ----------------------------------------------------------------

Add an H2 Overview header titled "Overview" or something like that.

The first section is in bold when It

probably shouldn't be.

However the

**It is always recommended section

isn't in bold (One side has 2 asterisks and

one has 3, should be 2 on both sides)

I think there's a spelling error in the

LangChain block ReAct Agen →

ReAct Agent

You might be able to embed that Youtube

video in the markdown with this code:

<iframe width="560" height="315" src="https://www.youtube-nocookie.com/embed/fqNBNE_YSro?si=JKaKOWfzdxeFFADe" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" allowfullscreen></iframe>

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

holtskinner commented on 2024-01-17T14:17:59Z ----------------------------------------------------------------

I appreciate the enthusiasm, but maybe

don't include the exclamation point.

Can you restructure this Installation

section to be like this notebook?

https://github.com/GoogleCloudPlatform/generative-ai/blob/main/gemini/getting-

started/intro

_gemini

_python.ipynb

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

holtskinner commented on 2024-01-17T14:18:00Z ----------------------------------------------------------------

Remove the output cells

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

holtskinner commented on 2024-01-17T14:18:01Z ----------------------------------------------------------------

Remove your own project ID, replace with <PROJECT ID>

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

holtskinner commented on 2024-01-17T14:18:02Z ----------------------------------------------------------------

Remove this output, it could be a security

risk.

There's existing auth instructions using

the gcloud SDK in this notebook

https://github.com/GoogleCloudPlatform/generative-ai/blob/main/language/getting-

started/intro_palm_api.ipynb

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

holtskinner commented on 2024-01-17T14:18:03Z ----------------------------------------------------------------

Move all library imports later in the

notebook to this cell

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

holtskinner commented on 2024-01-17T14:18:04Z ----------------------------------------------------------------

Make this header less verbose. Maybe

"Configure and call Healthcare API"?

Add the below bullet points about the API

into this section.

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

holtskinner commented on 2024-01-17T14:18:05Z ----------------------------------------------------------------

Where did this discharge summary come

from?

Do we have permission to use it?

Also, was it typed? From Speech to Text?

From OCR? Something else? Describe

that in the markdown.

I also don't think this needs to be in the notebook twice, I'd say just have it in the variable

elhadik commented on 2024-01-22T02:04:20Z ----------------------------------------------------------------

yes! it is the same one used by google healthcare API public website

https://cloud.google.com/healthcare-api/docs/how-tos/nlp-demo

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

holtskinner commented on 2024-01-17T14:18:06Z ----------------------------------------------------------------

Move the bullet points and image to the

top of this section. Also add a

link/description of "Custom Tool"

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

holtskinner commented on 2024-01-17T14:18:07Z ----------------------------------------------------------------

Can this be restructured to use the

Python discovery library if possible

instead of a raw REST call with the

requests library?

You can find code samples here

https://cloud.google.com/healthcare-api/docs/samples

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

holtskinner commented on 2024-01-17T14:18:08Z ----------------------------------------------------------------

Be sure to always capitalize "Healthcare

API"

Fix spelling Knowldge Graph

Capitalize Python

Add link to library networkx and put in code font

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

holtskinner commented on 2024-01-17T14:18:09Z ----------------------------------------------------------------

Once again, use the python discovery

library if possible.

And just a ease of reading change - Move

the function declarations to a separate cell

from the function calling.

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

holtskinner commented on 2024-01-17T14:18:10Z ----------------------------------------------------------------

Line #31.    response = healthcare_nl(TEXT)

Might be different if you switch to the

discovery version, but I think you can

move the .json() call to this line and

not have to do response.json()

throughout this cell

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

holtskinner commented on 2024-01-17T14:18:11Z ----------------------------------------------------------------

Line #33.    response.json()

Do Lines 32 and 33 accomplish anything?

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

holtskinner commented on 2024-01-17T14:18:12Z ----------------------------------------------------------------

Line #63.    kg_df = pd.DataFrame({"source": Source, "target": Target, "edge": Edge})

Here's another way to write this cell that's a little more efficient/readable. Be sure to run/test it before committing.

# Extract the entities from the response
response_json = healthcare_nl(TEXT).json()
Create DataFrame for entities
df = pd.DataFrame.from_records([
{
"ID": entity["mentionId"],
"Desc": f"{entity['type']} : {entity['text']['content']}",
"Type": entity["type"],
"Content": entity["text"]["content"],
}
for entity in response_json["entityMentions"]
])
Create DataFrame for knowledge graph
kg_df = pd.DataFrame.from_records([
{
"source": df.loc[df["ID"] == relation["subjectId"], "Desc"].values[0],
"target": df.loc[df["ID"] == relation["objectId"], "Content"].values[0],
"edge": df.loc[df["ID"] == relation["objectId"], "Type"].values[0],
}
for relation in response_json["relationships"]
])
Create a directed graph from the DataFrame
G = nx.from_pandas_edgelist(
kg_df, "source", "target", edge_attr=True, create_using=nx.MultiDiGraph()
)

elhadik commented on 2024-01-22T02:40:14Z ----------------------------------------------------------------

worked as well! big thanks

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

holtskinner commented on 2024-01-17T14:18:13Z ----------------------------------------------------------------

Create a new header that describes this

section.

capitalize LangChain & Healthcare API

Also add a hyperlink to a page about

ReAct

Fix the extra ** characters at the end

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

holtskinner commented on 2024-01-17T14:18:14Z ----------------------------------------------------------------

Fix the deprecation warning and use

langchain-google-vertexai

Also, looks like there's multiple printed

outputs, can this be made more concise?

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

holtskinner commented on 2024-01-17T14:18:15Z ----------------------------------------------------------------

Line #6.    from langchain.tools import StructuredTool

Add these imports at the beginning of the

notebook

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

holtskinner commented on 2024-01-17T14:18:16Z ----------------------------------------------------------------

Line #12.    tool = StructuredTool.from_function(list_of_medical_terms)

I'd say just create tools here instead of creating a tool variable

tools =  [StructuredTool.from_function(list_of_medical_terms)]

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

holtskinner commented on 2024-01-17T14:18:17Z ----------------------------------------------------------------

Is this block performing the self-

consistency check?

Can you add more to the description

outlining what the code block is doing?

The code can be simplified in this way

(Please double check before committing)

from collections import Counter # Add import to the top of the notebook
num_attempts = 10
outputs = [agent.run(prompt) for _ in range(num_attempts)]
final_output contains the most common output, and vote contains its count
final_output, vote = Counter(outputs).most_common(1)[0]

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

holtskinner commented on 2024-01-17T14:18:54Z ----------------------------------------------------------------

Can you add a description of this code block and how it's different than the first block?

[elhadik]

yes! it is the same one used by google healthcare API public website

https://cloud.google.com/healthcare-api/docs/how-tos/nlp-demo

---

View entire conversation on ReviewNB

[elhadik]

worked as well! big thanks

---

View entire conversation on ReviewNB