Chat

Scout Chat provides an AI-powered interface for natural language querying of the Scout data lake. Ask questions in plain English and receive data-driven answers with direct access to your radiology report data.

Note

Chat is optional and may not be enabled in all deployments. If you don’t see Chat on the Launchpad, contact your administrator.

Current version: Scout Chat queries HL7 radiology report data. Future versions will support DICOM metadata, pathology reports, and extracted features.

Overview

Scout Chat is powered by Open WebUI with Ollama:

• Natural language SQL: Converts questions into SQL queries against the data lake

• Scout Explorer model: Custom-configured LLM that understands the Scout data schema

• Real-time data access: Direct Trino connection via MCP (Model Context Protocol)

• Context-aware responses: Understands Scout terminology, fields, and data structure

Getting Started

1. Navigate to the Scout Launchpad

2. Click the Chat card

3. Type your question in plain English

4. Press Enter to submit

5. The AI queries the database and provides an answer

Example Queries

Research & Cohort Identification

• How many patients have both a chest CT and a lung nodule diagnosis?

• Find all patients with MRI reports mentioning "multiple sclerosis" in the findings

• What's the age and sex distribution for patients with pneumonia diagnoses?

• List unique patients with both CT and PET scans in 2024

Operational & Trend Analysis

• How does CT volume compare this month vs last month?

• What's the average turnaround time by modality?

• Show me report volumes by month for the past year

Text Search & Exploration

• Find MRI reports mentioning "metastasis" in the findings section

• How many reports contain "incidental" in the impression?

• Search for chest X-rays with "opacity" in the findings from the last 6 months

Understanding Chat Responses

When you ask a question, Scout Chat:

1. Interprets your question in its “Thinking” mode

2. Executes a SQL query via the Trino MCP tool (expandable to view)

3. Retrieves results from the Scout data lake

4. Analyzes the data and provides a natural language answer

Viewing the SQL Query

Click the expandable View Result from scout-db_execute_query section to see the exact SQL query.

This is useful for:

• Understanding how the AI interpreted your question

• Learning SQL syntax for use in Analytics SQL Lab

• Debugging unexpected results

• Adapting queries for Notebooks

Tips for Effective Queries

Be Specific

❌ Show me pneumonia cases
✓ How many patients have pneumonia mentioned in the impression, grouped by age decade?

Use Scout Terminology

The AI understands the Scout data schema. Reference field names when relevant:

• Modality: CT, MRI, X-ray, US, NM, PET, etc.

• Report sections: impression, findings, addendum, technician note

• Demographics: age, sex, race, zip code

• Temporal: observation date, message date, turnaround time

• Clinical: diagnosis codes, service name, study instance UID

Ask Follow-up Questions

Scout Chat maintains conversation context:

User: How many patients have "pulmonary embolism" in the impression?
Chat: There are 1,234 unique patients with pulmonary embolism mentioned.

User: What's the age distribution?
Chat: [Shows breakdown by age group]

User: Filter to just CT angiography studies
Chat: [Shows 892 patients with CTA studies mentioning PE]

Specify Date Ranges

How many reports from January 2024 to December 2024?
Show me the number of X-rays in the last 6 months

Request Tabular Data

Give me a table of report counts by modality, sorted highest to lowest
List the top 10 diagnosis codes with their counts

For visualizations, copy results to Analytics.

Data Privacy and Security

• Authentication required: Keycloak authentication (same as other Scout services)

• Read-only access: Chat cannot modify or delete data

• External content blocked: Scout blocks loading images and resources from external websites

• Conversation privacy: Chat history is stored on the server and associated with your user account. Other users cannot see your chats unless you share them.

Note

Admin visibility: Scout administrators have the ability to view user chat histories for quality assurance and support. Avoid including sensitive personal information in your conversations.

External Images and Links

Scout Chat includes security protections that block external content. If the AI generates a response containing an image from an external service (such as a charting website), the image will not render.

Warning

Do not click links to external websites in chat responses.

LLM responses may contain links to third-party services. These links could potentially contain sensitive data from your query embedded in the URL. If you see a broken image or an external link, do not click it.

For visualizations, copy the data to Analytics and build charts there.

Chat Sharing

You can share chat conversations with other authenticated Scout users via share links.

Warning

PHI Risk: Chat conversations may contain Protected Health Information (PHI) from query results. Before sharing or downloading chats, ensure you are complying with your institution’s data governance policies and HIPAA requirements.

Creating a Share Link

1. Open the chat you want to share

2. Click the three-dot menu (⋮) on the chat

3. Select Share

4. Click Copy Link to generate a shareable URL

The link creates a snapshot of the conversation at that moment. New messages added after sharing won’t appear unless you update the link.

Who Can View Shared Chats

Shared chats are only accessible to authenticated users on your Scout instance. Recipients must:

• Have a Scout account

• Be logged in to Scout Chat

Unauthenticated users will be redirected to the login page.

Updating a Share Link

If you add messages to a shared chat and want to include them:

1. Open the chat and click the three-dot menu

2. Select Share

3. The share modal shows the previously shared snapshot

4. Click Update to refresh the snapshot with new messages

Deleting a Share Link

To revoke access to a shared chat:

1. Open the chat and click the three-dot menu

2. Select Share

3. Click Delete this link

Once deleted, the share link becomes invalid and viewers can no longer access the chat.

Downloading Chats

Warning

Do not download chats containing PHI unless you have appropriate authorization and secure storage. Downloaded chat files may contain patient identifiers, diagnosis codes, and other sensitive information extracted from query results.

Because of PHI concerns, we recommend using chat sharing instead of download when you can. If this is not suitable for your use case, consider whether Scout Analytics or Scout Notebooks would meet your need.

Limitations

Data Scope

Scout Chat queries data in the Scout data lake only.

Current version:

• HL7 radiology report data only

• No PACS image access (DICOM support planned)

• No external database queries

Query Complexity

For advanced analysis, consider:

• Analytics: Persistent visualizations, dashboards, and complex SQL

• Notebooks: Statistical analysis, machine learning, and custom transformations

Model Limitations

The AI may occasionally misinterpret questions or generate incorrect queries. Always review the SQL in the View Result from scout-db_execute_query section to verify it matches your intent.

Troubleshooting

Chat Service Not Available

If Chat doesn’t appear on the Launchpad, the service may not be enabled in your deployment. Contact your Scout administrator.

No Response or Errors

1. Be patient — GPU resources may be limited with concurrent users

2. Retry — Occasionally the model makes formatting errors

3. Log out and back in — Refreshes your session

4. Contact admin — If issues persist

Unexpected Results

1. Expand View Result from scout-db_execute_query to review the SQL query

2. Verify your question was specific and unambiguous

3. Check if the data contains what you expect

4. Rephrase with more specific criteria

Tool Not Working

If you see Trino tool errors, contact your administrator to verify the MCP tool configuration.

Additional Resources

• Data Schema: Available fields and their meanings

• Services Overview: Analytics, Notebooks, and other Scout services

• Tips & Tricks: General Scout usage tips