RadGPT

<!--
# COPYRIGHT NOTICE
# This file is part of the "Universal AI Agentic Skills" project.
# Copyright (c) 2026 MD BABU MIA, PhD <md.babu.mia@mssm.edu>
# All Rights Reserved.
#
# This code is proprietary and confidential.
# Unauthorized copying of this file, via any medium is strictly prohibited.
#
# Provenance: Authenticated by MD BABU MIA

-->

---
name: radgpt-radiology-reporter
description: Radiology Reporter
keywords:
  - radiology
  - report-generation
  - patient-friendly
  - summarization
  - explanation
measurable_outcome: Generate a patient-friendly explanation of a radiology report with <1% hallucination rate within 30 seconds.
license: MIT
metadata:
  version: "1.0.0"
compatibility:
  - system: Python 3.9+
allowed-tools:
  - run_shell_command
  - read_file
---

# RadGPT (Radiology Report Assistant)

An LLM-based agent designed to summarize and explain complex radiology reports for patients and clinicians.

## When to Use

*   **Patient Communication**: Converting technical findings into plain language.
*   **Clinician Review**: Highlighting critical findings (e.g., "Pneumothorax detected").
*   **Follow-up**: Suggesting appropriate next steps based on findings.

## Core Capabilities

1.  **Simplification**: Translates "bilateral opacity" to "cloudiness in both lungs".
2.  **Entity Extraction**: Identifies key anatomical structures and pathologies.
3.  **Q&A**: Answers follow-up questions about the report.
4.  **MRI Report Patient Education**: Translates MRI reports into personalized plain language while preserving diagnostic meaning, uncertainty, negation, laterality, severity, follow-up recommendations, and red flags; sets and checks an audience-appropriate readability target; checks the explanation against the source report for omissions and hallucinations; compares consequential interpretations with a qualified radiology or clinical expert; and explicitly separates patient education from diagnosis or treatment advice, escalating consequential findings for clinician review before delivery.
5.  **Closed-System Radiology-Response Evaluation**: Evaluate closed-system radiology responses for correctness, relevance, omissions, unsafe recommendations, uncertainty communication, and agreement with authoritative radiography references.
6.  **Closed-System LLM Radiography Checklist**: Review AI-generated radiography responses from closed-system LLMs for factual correctness, clinical risk, uncertainty language, scope limits, and comparison against applicable radiography education or reporting standards.
7.  **Closed-System Radiography Response Safeguards**: Assess locked or text-only LLM radiography answers for factual correctness, omitted safety context, patient-facing language quality, escalation triggers for clinician or radiologist review, and stated limits when used in radiologic education workflows.
8.  **Closed-System LLM Radiography Response Validation**: Log prompt text and model version for each closed-system LLM radiography response; compare outputs with applicable radiography educational standards; detect unsafe omissions and overstatements; and route responses with clinical uncertainty, omitted safety context, or potential patient-care impact to radiologist review.
9.  **Closed-System Radiography-Response QA**: Check closed-system LLM radiography outputs for technical accuracy, patient-facing readability, stated scope limits, appropriate escalation language, source grounding in the provided report or authoritative references, and review by radiography or radiology professionals before clinical use.
10. **Closed-System LLM Radiography Response Evaluation**: Build modality-specific radiography prompt sets; have radiographers review response correctness; check patient-facing language for clarity and safety; capture hallucinations, unsupported claims, and missing caveats; and document limits when image context is absent or unavailable.
11. **Closed-System LLM Radiography Response QA Mode**: Assess closed-system LLM outputs for radiographic positioning and projection terminology, anatomy-specific correctness, omitted safety context, and clear separation of patient-facing explanations from diagnostic claims requiring qualified clinical review.
12. **Closed-System LLM Radiography Source-Finding Evaluation**: Check AI-generated radiography responses for correct radiography terminology, factual alignment with the provided source findings, patient-facing readability, hallucinated imaging findings, and clear technologist/radiologist review boundaries before clinical use.

## Workflow

1.  **Input**: Raw text of the radiology report.
2.  **Process**: LLM summarizes and identifies key findings.
3.  **Output**: Structured summary or conversational explanation.

## Example Usage

**User**: "Explain this chest X-ray report to the patient."

**Agent Action**:
```bash
python -m radgpt.explain --report ./report.txt --target_audience patient
```

## References

*   https://pubmed.ncbi.nlm.nih.gov/41865475/
*   https://pubmed.ncbi.nlm.nih.gov/42024724/


<!-- AUTHOR_SIGNATURE: 9a7f3c2e-MD-BABU-MIA-2026-MSSM-SECURE -->