Image Based AI Workflow - Brief Proposal

From IHE Wiki
Jump to navigation Jump to search

IHE Work Item Proposal (Short)

1. Proposed Work Item: Imaged- Based AI Workflow

Proposal Editor: Christopher Lindop, Shujah Das Gupta

Work item Editor: Chris Lindop

Date: 10th August 2018/update 21 August 2018

Version: V 1.1

Domain: Radiology

2. The Problem

With the introduction of AI (Artificial Intelligence) based actors into the Radiology workflow, there is a significant challenge in integrating the workflow steps with existing Radiology actors (as outline outlined in the SWF, CDS-OAT, and PAW) and defining the transactions which need to be supported for the workflow.

The current approach being taken by most vendors is focused on introducing proprietary transactions and integrations which lead to a significantly fragmented workflow and introduce multiple new interoperability challenges. The proposed IHE work-item will outline well-defined transactions to support the integration of AI based actors leveraging existing standards-based operations to support the workflow and facilitate interoperability across these actors

3. Key Use Case

There are 3 primary workflows which could be addressed using existing standards for the integration of AI workflows:

1. Scheduling/triggering an AI algorithm for a specific case

2. Providing an update once the AI algorithm has been invoked and communicating the update to the billing system

3. Creating artifact(s) which are based on existing standards that can be linked to the patient record permanently

An example use case is CT lung nodule AI analysis (potentially leveraging PAWF workflow) to trigger human worklist prioritization.

  1. Acquisition Modality(CT) acquires and archives CT Lung study to PACS (Image Manager/Archive)
  2. PACS(Image Manager/Archive) notifies the Radiology Workflow Manager(Workitem Manager) that CT Images are available for Lung Node AI Analysis
  3. Radiology Workflow Manager(Workitem Manager) Creates workitem for the AI Server(Workitem performer) to act on.
  4. AI Server (Workiten Performer) accepts the CT Lung Node AI Analysis workitem, retrieves the CT images and begins the AI Analysis task.
  5. AI Server (Workitem Performer) completes the CT Lung Node AI Analysis workitem, stores the Analysis results and notifies the Radiology Workflow Manager of success with a prioritization score for human follow-up.
  6. Radiology Workflow Manager receives the completed CT Lung Node AI Analysis workitem, with the location of the analysis results and a prioritization score for human follow-up.
  7. Radiology Workflow Manager uses the prioritization score for modifying the Radiologist worklist reading priority as appropriate for followup.

4. Standards & Systems

The proposal will look to leverage existing workflows and standards to support the integration of these actors through the following:

1. Leverage the transactions supported in the Radiology Workflow profiles (WSF.b and PAW)

2. Leverage existing IHE Workflow transactions for communication with the Healthcare IT systems for scheduling/billing/use tracking (CDS-OAT)

3. Defining standardized artifacts for the output of AI Algorithms (DICOM SR, Leveraging the SR IOD provided in PS3.21 to support transformations with other representations with potential FHIR based transactions)

5. Discussion

IHE will be an appropriate forum to address the interoperability scenarios as the workflow steps will involve interactions across multiple actors and several transactions based on existing and developing standards can be leveraged to support the interactions.

This proposal is for a profile or whitepaper to facilitate adoption of the new as well as established DICOM and HL7 standards to address the challenges of AI in Image-based workflow.

The primary focus is to promote the use of existing standards. New standards development, if needed, is expected to be completed by the appropriate SDO.