FastPass - Utilizing OCR and Microsoft Text Analytics For Health to Create FHIR Data

Pre-requisite:

Before diving in, please read through Creating FHIR Objects Using Power Apps and FHIRPower a Custom Connector. This previous blog article highlights a low code version of the concepts we will address in the following article.

Another key component of FastPass is Microsoft's Text Analytics for Health cognitive service. David on the HLS Emerging Opportunities Team has written a fantastic article delving into the Text Analytics for Health Use Cases.

Beyond that there will be an emphasis on Azure Functions, Azure Static Web Apps, DOTNET version 7, and Azure API for FHIR.

Why:

We built this project over two different phases, each representing two different possible operations of health care data.

Medical Intake Forms - Patient presents medical insurance card at the doctor's office. An image of the insurance card is uploaded to the system and sent to the OCR service to extract patient biographic information as well as insurance coverage information.

The text extracted by the OCR is presented on a form for a intake specialist to verify, correct, and then submit to the EMR system using the FHIR API exposed by the Azure Health Data Service.

Discharge Summary Analysis - For a given patient, we obtain a discharge summary and extract relevant medical information. Using a custom web application, we submit the contents of the discard summary to the Text Analytics for Healthcare service.

Upon successful processing, we extract the entities identified as well as the FHIR bundle. The extracted entities are then displayed on a web page on a custom web application to conduct quick spot/quality check. Once the data is validated, the extracted FHIR bundle is submitted to the Azure Health Data Services for saving into the EMR using the Azure Health Data Service FHIR API.

Solution Architecture Overview:

Solution Architecture Diagram

We built a custom web application using Blazor to handle the application flow. The application manages the two use cases mentioned above, the medical intake as well as the discharge summary analysis.

The application will be deployed as a static website for performance purposes. We want to improve the performance, while also lowering the deployment cost. Since we are using a static website for production deployment, we’ll need to using the SWA CLI for local development. When you deploy a static website on Azure, Azure provides the backend support required to run the application. This includes wiring up support of Authentication/Authorization as well as access to Azure Functions. In order to support local development, this infrastructure is not available, which is what the SWA CLI provides.

More information on running the project locally can be found here in the Setup Guide section.

Conclusion:

FastPass can be a heavy lifter in terms of creating and maintaining patient records because it automates much of the typical workflow, while allowing for critical human oversight.

The project is open source because we see the potential for growth spear headed by the wider community. We at the HLS Emerging Opportunities team would love to see feedback and any potential contributions to FastPass.

Continue reading...