agentleFS
aizechaizech/clinical-skills/.agents/skills/llm-radiology-use/SKILL.md

llm-radiology-use

Skill4 starsChanged 7 months agoMITView on GitHub

Use LLM APIs for radiology tasks. Also use when integrating medical LLMs (MedPaLM, MedLM, Google Health, Amazon HealthLake) for report analysis, clinical reasoning, or radiology AI workflows.

The file

We split this skill into the rules it gives an agent

2 rules. About 0 are common, 2 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Validate outputs: - Always review LLM-generated content

    line 314
  2. Distinctive:

    Monitor for hallucinations: - Verify against source data

    line 318

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

LLM for Radiology

You are an expert in medical large language models (LLMs) for radiology applications. Your role is to help users integrate and optimize LLM-based radiology workflows.

Supported LLM Platforms

PlatformFocusCapabilities
MedPaLM/MedLMMedical reasoningReport analysis, QA
Google HealthMedical imagingMulti-modal reasoning
Amazon HealthLakeHealthcare dataFHIR integration
Azure AI HealthMedical NLPClinical insights
Claude HealthMedical reasoningReport analysis

Key Concepts

Medical LLM Capabilities

  • Report summarization
  • Finding extraction
  • Clinical reasoning
  • Prior study comparison
  • Structured data extraction
  • Quality assessment

Prompt Engineering

SYSTEM_PROMPT = """You are an expert radiologist assistant. 
Your role is to analyze radiology reports and provide insights.
Always be clinically accurate and evidence-based.
Prioritize patient safety in all recommendations."""

MedPaLM Integration

API Configuration

import requests
import json

MEDPALM_API = "https://generativelanguage.googleapis.com/v1beta1"

def configure_medpalm(api_key):
    """Configure MedPaLM API."""
    return {
        "base_url": MEDPALM_API,
        "api_key": api_key,
        "model": "medpalm-2"
    }

def query_medpalm(config, prompt, context=None):
    """Query MedPaLM for radiology insights."""
    url = f"{config['base_url']}/models/{config['model']}:generateContent"
    
    contents = [{"parts": [{"text": prompt}]}]
    
    if context:
        contents[0]["parts"][0]["text"] = f"Context: {context}\n\nQuestion: {prompt}"
    
    response = requests.post(
        f"{url}?key={config['api_key']}",
        headers={"Content-Type": "application/json"},
        json={
            "contents": contents,
            "generationConfig": {
                "temperature": 0.2,
                "topP": 0.8,
                "maxOutputTokens": 1024
            }
        }
    )
    
    return response.json()

Report Analysis Prompt

REPORT_ANALYSIS_PROMPT = """Analyze the following radiology report and provide:
1. Key findings summary
2. Critical findings (if any)
3. Clinical recommendations
4. Suggested follow-up

Report:
{report_text}

Respond in structured format."""

def analyze_report(config, report_text):
    """Analyze radiology report with MedPaLM."""
    prompt = REPORT_ANALYSIS_PROMPT.format(report_text=report_text)
    return query_medpalm(config, prompt)

Google Health Integration

Medical Imaging API

GOOGLE_HEALTH_API = "https://health.googleapis.com/v1"

def configure_google_health(credentials_path):
    """Configure Google Health API."""
    return {
        "base_url": GOOGLE_HEALTH_API,
        "credentials": credentials_path
    }

def medical_insights(config, study_data):
    """Get medical imaging insights."""
    response = requests.post(
        f"{config['base_url']}/projects/{config['project']}/locations:improve",
        headers={"Authorization": f"Bearer {get_token(config)}"},
        json=study_data
    )
    return response.json()

Amazon HealthLake Integration

FHIR-Based Integration

import boto3

def configure_healthlake(region="us-east-1"):
    """Configure Amazon HealthLake."""
    return {
        "client": boto3.client("healthlake", region_name=region),
        "datastore_id": None  # Set after creation
    }

def query_imaging_history(config, patient_id):
    """Query patient imaging history from HealthLake."""
    response = config["client"].search_by_range(
        AssetId=patient_id,
        SearchParameters={
            "filters": {
                "DocumentType": {"Value": "DiagnosticReport", "Type": "String"}
            }
        }
    )
    return response["Results"]

Send Imaging Results

def send_results_to_healthlake(config, patient_id, report_data):
    """Send radiology report to HealthLake."""
    config["client"].create_fhir_resource({
        "ResourceType": "DiagnosticReport",
        "subject": {"reference": f"Patient/{patient_id}"},
        "status": "final",
        "code": {"text": report_data["study_type"]},
        "conclusion": report_data["impression"]
    })

Azure AI Health

Health NLP Configuration

AZURE_ENDPOINT = "https://<resource>.cognitiveservices.azure.com"

def configure_azure_health(endpoint, api_key):
    """Configure Azure AI Health."""
    return {
        "endpoint": endpoint,
        "api_key": api_key
    }

def extract_medical_entities(config, text):
    """Extract medical entities from report."""
    response = requests.post(
        f"{config['endpoint']}/text/analytics/v3.1/entities/health",
        headers={
            "Ocp-Apim-Subscription-Key": config["api_key"],
            "Content-Type": "application/json"
        },
        json={"documents": [{"id": "1", "text": text}]}
    )
    return response.json()

Structured Data Extraction

Report to Structured Format

EXTRACTION_PROMPT = """Extract structured data from this radiology report:

Report: {report_text}

Extract and format as JSON:
{{
    "patient_id": "...",
    "study_type": "...",
    "findings": [
        {{
            "anatomy": "...",
            "finding": "...",
            "size": "...",
            "location": "..."
        }}
    ],
    "impression": "...",
    "critical_findings": [...],
    "recommendations": [...]
}}"""

def extract_structured(config, report_text):
    """Extract structured data from report."""
    prompt = EXTRACTION_PROMPT.format(report_text=report_text)
    response = query_llm(config, prompt)
    
    # Parse JSON from response
    return json.loads(extract_json(response))

Clinical Reasoning

Comparison Analysis

COMPARISON_PROMPT = """Compare these two CT reports and identify changes:

Current Report:
{current}

Prior Report:
{prior}

Identify:
1. New findings
2. Resolved findings
3. Changed findings (with details)
4. Stable findings
5. Clinical significance"""

def compare_reports(config, current, prior):
    """Compare current and prior reports."""
    prompt = COMPARISON_PROMPT.format(current=current, prior=prior)
    return query_llm(config, prompt)

Differential Diagnosis

DIFFERENTIAL_PROMPT = """Based on these imaging findings, provide differential diagnosis:

Findings: {findings}
Modality: {modality}
Clinical history: {history}

For each differential:
1. Diagnosis
2. Key supporting features
3. Most likely ranking
4. Recommended additional imaging (if needed)"""

def get_differential(config, findings, modality, history):
    """Get differential diagnosis."""
    prompt = DIFFERENTIAL_PROMPT.format(
        findings=findings,
        modality=modality,
        history=history
    )
    return query_llm(config, prompt)

Batch Processing

Bulk Report Analysis

def batch_analyze_reports(config, reports, batch_size=10):
    """Analyze multiple reports in batch."""
    results = []
    
    for i in range(0, len(reports), batch_size):
        batch = reports[i:i + batch_size]
        batch_results = []
        
        for report in batch:
            try:
                result = analyze_report(config, report["text"])
                batch_results.append({
                    "report_id": report["id"],
                    "analysis": result
                })
            except Exception as e:
                batch_results.append({
                    "report_id": report["id"],
                    "error": str(e)
                })
        
        results.extend(batch_results)
    
    return results

Best Practices

  1. Validate outputs - Always review LLM-generated content
  2. Use appropriate temperature - Lower (0.2-0.3) for factual analysis
  3. Provide context - Include clinical history when available
  4. Set confidence thresholds - Flag low-confidence responses
  5. Monitor for hallucinations - Verify against source data

Troubleshooting

IssueSolution
Slow responsesUse batch processing
Inaccurate outputRefine prompt with examples
Missing dataEnsure context is complete
Rate limitsImplement backoff strategy

Related Skills

  • radiology-report-analysis: For report analysis basics
  • llm-radiology-use: For LLM integration (this skill)
  • ai-quality-review: For output validation
  • guideline-integration: For evidence-based recommendations

Examples

Example 1: Analyze Report

Use MedPaLM to analyze this chest CT report
report = "CT CHEST: 2.5cm mass right upper lobe..."
analysis = analyze_report(config, report)

Example 2: Compare Studies

Compare this CT with the prior study from 3 months ago
comparison = compare_reports(
    config,
    current="Current report text...",
    prior="Prior report text..."
)

Example 3: Extract Structured Data

Extract findings from this report to structured format
structured = extract_structured(config, report_text)
# Returns JSON with findings, measurements, etc.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.