Cardiology Visual System

A unified meta-skill that intelligently routes visual content requests to the optimal tool. No more switching between Napkin.ai, NotebookLM, or other subscriptions—everything happens in Claude Code.

Quick Reference: What Tool Does What

You Ask For	Tool Used	Output
Blog header, lifestyle photo, patient scenario	Fal.ai	PNG image
Infographic, explainer graphic, medical illustration	Gemini	PNG/JPG image
Flowchart, treatment algorithm, clinical pathway	Mermaid	SVG/PNG diagram
Slide deck, presentation	Marp	PPTX/PDF/HTML slides
Data chart, trial results, trends over time	Plotly	Interactive HTML or PNG
Interactive explainer, dashboard	React Artifact	Interactive HTML
Quick visualization prototype, exploratory data viz	LIDA ⚠️	PNG + Code (prototype only)

Automatic Routing Logic

When you ask for visuals, I determine the best tool by analyzing your request:

→ Route to Fal.ai (Stock/Human Imagery)

Keywords: blog image, header, hero image, lifestyle, patient photo, stock, person, family, emotional, scenario

Best for:

Blog post headers
Patient experience illustrations
Lifestyle/wellness imagery
Emotional/human-centered scenes
Recovery and hope imagery

NOT for: Medical devices, ECGs, diagrams, data, or text-heavy content

→ Route to Gemini (Infographics & Medical Illustrations)

Keywords: infographic, explainer, illustration, visual summary, concept diagram, icons, steps, process visual, simplified, educational graphic

Best for:

Infographics (like Napkin.ai produces)
Medical concept illustrations
Simplified explainer graphics
Educational visuals with icons
Text-in-image content
Visual summaries of articles

→ Route to Mermaid (Diagrams & Flowcharts)

Keywords: flowchart, algorithm, pathway, decision tree, sequence, timeline, process flow, treatment algorithm, diagnostic pathway, workflow

Best for:

Clinical decision trees
Treatment algorithms
Diagnostic pathways
Process workflows
Organizational charts
Sequence diagrams
Gantt charts for timelines

→ Route to Marp (Slide Decks)

Keywords: slides, presentation, deck, powerpoint, lecture, talk, keynote

Best for:

Conference presentations
Educational lectures
Patient education slides
Grand rounds presentations
CME content

→ Route to Plotly (Data Visualization)

Keywords: chart, graph, plot, data, statistics, trial results, forest plot, trends, comparison, survival curve, Kaplan-Meier, bar chart, line graph, scatter

Best for:

Clinical trial results
Statistical comparisons
Trends over time
Forest plots
Survival curves
Before/after data
Multi-study comparisons

→ Route to React Artifact (Interactive)

Keywords: interactive, dashboard, calculator, tool, widget, animated, explorable

Best for:

Risk calculators
Interactive explainers
Animated diagrams
Patient education tools
Explorable explanations

→ Route to LIDA (Quick Prototyping) ⚠️ PROTOTYPING ONLY

Keywords: quick, prototype, exploratory, rough draft, multiple options, try, experiment, brainstorm visualization

⚠️ CRITICAL LIMITATIONS:

PROTOTYPING ONLY - NOT for publication or patient-facing materials
Quality varies - ALWAYS review for medical accuracy
Works best with ≤10 data columns
No specialized medical charts (true forest plots, Kaplan-Meier)
Requires manual validation before ANY use

Best for:

Quick exploratory data visualization ("what does this data show?")
Generating multiple visualization candidates
Brainstorming chart types for new data
Internal research review only

NOT for:

Publication-ready charts → Use Plotly instead
Patient-facing materials → Use production tools
Regulatory submissions → Never
Final blog posts → Use Plotly or Gemini

When in doubt, use Plotly for data visualization instead of LIDA.

Tool 1: Fal.ai (Blog Imagery)

Setup

export FAL_KEY="your-fal-api-key"
# Get key from: https://fal.ai/dashboard/keys

Usage

python scripts/fal_image.py "A 55-year-old man experiencing chest pain at work" --output hero.png

Models

Model	Cost	Best For
`fal-ai/recraft-v3`	$0.04	Default - Best quality
`fal-ai/flux-pro/v1.1`	$0.04	Photorealism
`fal-ai/flux/schnell`	$0.003	Fast/cheap drafts

What to Generate vs Not Generate

✅ GENERATE:

Patient symptoms/experiences (chest pain, shortness of breath)
Lifestyle scenes (exercise, healthy cooking)
Doctor-patient conversations
Family support moments
Recovery celebrations

❌ DO NOT GENERATE:

Medical devices (pacemakers, stents, valves)
Clinical imagery (ECGs, angiograms, OR scenes)
Anatomical diagrams
Medications

Tool 2: Gemini (Infographics)

Setup

export GEMINI_API_KEY="your-gemini-api-key"

Usage

python scripts/gemini_infographic.py \
  --topic "Heart Failure Stages" \
  --style "minimalist medical" \
  --output hf_stages.jpg

Prompting for Medical Infographics

Structure your prompt:

Create a [STYLE] infographic showing [TOPIC].

Include:
- [Key point 1]
- [Key point 2]
- [Key point 3]

Style: [clean/minimalist/modern medical], use icons, clear hierarchy, 
professional color palette (blues, teals for medical), easy to read text

Example prompts:

Disease Progression:

"Create a minimalist medical infographic showing the 4 stages of heart failure (A, B, C, D). Use icons for each stage, show progression with arrows, include brief descriptions. Clean layout, medical blue color scheme."
Treatment Comparison:

"Create an infographic comparing medication vs intervention for AFib. Two columns, icons for each approach, bullet points for pros/cons. Modern medical style."
Risk Factor Summary:

"Create a visual summary of 7 modifiable risk factors for heart disease. Icon for each factor, clean grid layout, actionable tips. Professional medical illustration style."

Tool 3: Mermaid (Diagrams)

You have Mermaid Chart MCP connected. Use it for structured diagrams.

Common Cardiology Diagram Types

1. Treatment Algorithm:

flowchart TD
    A[Acute Chest Pain] --> B{STEMI?}
    B -->|Yes| C[Primary PCI < 90 min]
    B -->|No| D{High-risk NSTEMI?}
    D -->|Yes| E[Early invasive < 24h]
    D -->|No| F[Ischemia-guided strategy]

2. Clinical Pathway:

flowchart LR
    A[Diagnosis] --> B[Risk Stratification]
    B --> C[Treatment Selection]
    C --> D[Follow-up Protocol]

3. Diagnostic Decision Tree:

flowchart TD
    A[Dyspnea] --> B{BNP elevated?}
    B -->|Yes| C{Echo findings?}
    B -->|No| D[Consider other causes]
    C -->|HFrEF| E[GDMT initiation]
    C -->|HFpEF| F[Diuretics + address comorbidities]

4. Timeline (Gantt):

gantt
    title Post-MI Care Timeline
    dateFormat  YYYY-MM-DD
    section Acute
    Hospital stay        :a1, 2024-01-01, 5d
    section Recovery
    Cardiac rehab        :a2, after a1, 12w
    section Long-term
    Medication titration :a3, after a1, 6m

Tool 4: Marp (Slides)

Setup

npm install -g @marp-team/marp-cli

Usage

I write Markdown with Marp syntax
Save as presentation.md
Convert:

marp presentation.md --pptx           # PowerPoint
marp presentation.md --pdf            # PDF
marp presentation.md -o slides.html   # HTML

Marp Template for Medical Slides

---
marp: true
theme: default
paginate: true
backgroundColor: #ffffff
color: #333333
---

# Heart Failure Management
## Modern Approaches in 2024

Dr. [Your Name]
Interventional Cardiology

---

# Agenda

1. Current Guidelines
2. New Therapies
3. Case Discussion

---

# Key Statistics

- 6.7 million Americans with HF
- 50% mortality at 5 years
- $30.7 billion annual cost

![bg right:40%](path/to/chart.png)

---

# Treatment Algorithm

```mermaid
flowchart TD
    A[HFrEF Diagnosis] --> B[GDMT Initiation]
    B --> C[Titrate to target doses]

Take-Home Points

Early initiation matters
Quadruple therapy is standard
Device therapy in appropriate patients

Questions?

Contact: your@email.com


---

## Tool 5: Plotly (Data Visualization)

### Setup
```bash
pip install plotly kaleido pandas --break-system-packages

Common Medical Visualizations

1. Bar Chart (Trial Results):

import plotly.express as px

data = {
    'Treatment': ['Drug A', 'Drug B', 'Placebo'],
    'Event Rate (%)': [12.3, 15.1, 18.7]
}
fig = px.bar(data, x='Treatment', y='Event Rate (%)',
             title='Primary Endpoint: Major Cardiovascular Events',
             color='Treatment')
fig.write_html('trial_results.html')
fig.write_image('trial_results.png')

2. Forest Plot Style:

import plotly.graph_objects as go

studies = ['PARADIGM-HF', 'DAPA-HF', 'EMPEROR-Reduced']
hr = [0.80, 0.74, 0.75]
lower = [0.73, 0.65, 0.65]
upper = [0.87, 0.85, 0.86]

fig = go.Figure()
for i, study in enumerate(studies):
    fig.add_trace(go.Scatter(
        x=[lower[i], upper[i]], y=[study, study],
        mode='lines', line=dict(color='gray', width=2)
    ))
    fig.add_trace(go.Scatter(
        x=[hr[i]], y=[study],
        mode='markers', marker=dict(size=12, color='navy')
    ))

fig.add_vline(x=1.0, line_dash="dash", line_color="red")
fig.update_layout(title='Hazard Ratios for Heart Failure Trials',
                  xaxis_title='Hazard Ratio (95% CI)')

3. Trend Over Time:

import plotly.express as px

fig = px.line(df, x='Year', y='Mortality Rate', 
              color='Treatment Era',
              title='Heart Failure Mortality Trends 1990-2024')
fig.write_html('trends.html')

Tool 6: LIDA (Quick Prototyping) ⚠️ PROTOTYPING ONLY

⚠️ Critical Warning

LIDA is a PROTOTYPING TOOL ONLY. Do NOT use for:

Publication-ready visualizations
Patient-facing materials
Regulatory submissions
Final blog posts or social media

For production visualizations, use:

Plotly (data charts)
Gemini (infographics)
Fal.ai (images)

Setup

# Install LIDA
pip install lida llmx openai --break-system-packages

# Set API key (choose one)
export OPENAI_API_KEY="your-key"      # Recommended
export GOOGLE_API_KEY="your-key"      # Free tier (Gemini)
export ANTHROPIC_API_KEY="your-key"   # Claude

What LIDA Does

LIDA (Automatic Visualization Generation) uses LLMs to:

Analyze your data structure
Generate visualization code from natural language
Create multiple visualization candidates
Support multiple libraries (Plotly, Matplotlib, Seaborn, Altair)

Published: ACL 2023 (Microsoft Research)

Limitations

≤10 columns recommended - LLM context constraints
Quality varies - AI-generated, needs review
No specialized medical charts - No true forest plots, Kaplan-Meier
Error rate: <3.5% reported, but ALWAYS verify medical accuracy
Not production-ready - Use for exploration only

Usage

Basic Usage:

python scripts/lida_quick_viz.py "Show mortality by treatment group" trial_data.csv

Multiple Candidates:

python scripts/lida_quick_viz.py "Compare outcomes" data.csv --candidates 3

Use Medical Template:

python scripts/lida_quick_viz.py "Trial results" data.csv --template trial_comparison

Specify Library:

python scripts/lida_quick_viz.py "Trends over time" data.csv --library plotly

Interactive Mode:

python scripts/lida_quick_viz.py --interactive data.csv

List Templates:

python scripts/lida_quick_viz.py --list-templates

Medical Templates

Template	Description	Use Case
`trial_comparison`	Compare treatment arms	Primary endpoint results
`patient_demographics`	Baseline characteristics	Patient population summary
`outcome_comparison`	Primary/secondary endpoints	Multiple outcomes comparison
`trend_analysis`	Trends over time	Longitudinal data
`survival_curve`	Time-to-event (simplified)	Event-free survival (NOT true KM)

Example Workflow

1. Exploratory Analysis (LIDA):

# Quick exploration of new trial data
python scripts/lida_quick_viz.py \
  "Show primary endpoint by treatment arm with confidence intervals" \
  trial_results.csv \
  --template trial_comparison \
  --candidates 3

2. Review Candidates:

Check medical accuracy
Verify data interpretation
Select best approach

3. Production Version (Plotly):

# Recreate selected visualization in production quality
python scripts/plotly_charts.py bar --data trial_results.csv --output final_chart.png

Quality Validation Checklist

Every LIDA output includes this checklist:

⚠️  QUALITY VALIDATION CHECKLIST - REVIEW BEFORE USE

Medical Accuracy:
[ ] Data interpretation is correct
[ ] Statistical measures are appropriate
[ ] Confidence intervals/error bars are correct
[ ] P-values and significance are accurate
[ ] Sample sizes are represented correctly

Visual Design:
[ ] Chart type is appropriate
[ ] Color scheme is professional
[ ] Labels are clear and complete
[ ] Legend is accurate
[ ] Title describes the content

Medical Standards:
[ ] Follows publication standards
[ ] No misleading visualizations
[ ] Appropriate precision
[ ] Context is provided
[ ] Source attribution if needed

When to Use LIDA vs Plotly

Scenario	Tool	Reason
"What's the best way to show this data?"	LIDA	Exploration, multiple options
"Show me trial results quickly"	LIDA	Speed over perfection
"Generate 3 chart options"	LIDA	Multiple candidates
"Publication-ready chart"	Plotly	Production quality
"Blog post visualization"	Plotly	Final output
"Patient education"	Plotly	Accuracy critical
"Regulatory submission"	Plotly	Never use LIDA

Cost

Model	Cost	Speed	Quality
OpenAI (GPT-4o-mini)	$0.60/M tokens	Fast	Good
Gemini (Free tier)	FREE	Fast	Good
Claude (Sonnet)	Variable	Medium	Excellent

Typical cost per visualization: $0.01-0.05 (negligible)

Example: Interactive Session

python scripts/lida_quick_viz.py --interactive trial_data.csv

# In interactive mode:
> list                              # Show templates
> template trial_comparison         # Set template
> library plotly                    # Set library
> viz Show mortality by treatment   # Generate viz
> viz Compare age distribution      # Another viz
> quit

Output Structure

lida_output/
├── candidate_1_code.py           # Generated Python code
├── candidate_1.png                # Rendered visualization
├── candidate_2_code.py
├── candidate_2.png
└── ...

Integrated Workflow Example

Scenario: You're writing a blog post about heart failure medications.

Step 1: Hero Image (Fal.ai)

python scripts/fal_image.py \
  "Elderly patient having hopeful conversation with cardiologist about new treatment options" \
  --output images/hero.png

Step 2: Treatment Algorithm (Mermaid)

Create a Mermaid flowchart showing HFrEF GDMT initiation:
- Start with diagnosis
- Branch to ARNI/ACEi + Beta-blocker
- Add SGLT2i + MRA
- Device consideration

Step 3: Trial Data (Plotly)

# Compare mortality reduction across trials
trials_df = pd.DataFrame({
    'Trial': ['PARADIGM-HF', 'DAPA-HF', 'EMPEROR-Reduced'],
    'Mortality Reduction': [20, 17, 14]
})
fig = px.bar(trials_df, x='Trial', y='Mortality Reduction',
             title='Mortality Reduction in Landmark HF Trials (%)')

Step 4: Key Concepts Infographic (Gemini)

Create a clean medical infographic summarizing the "4 Pillars of HFrEF Therapy":
1. ARNI/ACEi - heart icon
2. Beta-blocker - heart rate icon  
3. MRA - kidney/electrolyte icon
4. SGLT2i - glucose/kidney icon

Style: modern medical, blue color scheme, minimal text, icon-focused

API Key Checklist

Before using this system, ensure these are set:

# Fal.ai (blog images)
export FAL_KEY="your-key"

# Gemini (infographics)
export GEMINI_API_KEY="your-key"

# LIDA (prototyping) - choose one
export OPENAI_API_KEY="your-key"      # Recommended
export GOOGLE_API_KEY="your-key"      # FREE (Gemini)
export ANTHROPIC_API_KEY="your-key"   # Claude

# Mermaid - uses MCP, no key needed
# Plotly - local, no key needed
# Marp - local, no key needed

Cost Summary

Tool	Cost	Typical Use
Fal.ai (Recraft)	$0.04/image	3-4 per blog = $0.16
Gemini	Free tier available	Infographics
LIDA (prototyping)	$0.01-0.05/viz	Quick exploration (optional)
Mermaid	Free (MCP)	Diagrams
Plotly	Free	Data viz
Marp	Free	Slides

Total per blog post: ~$0.16-0.25 (vs separate subscriptions)

Note: LIDA is optional - use FREE Gemini model for zero-cost prototyping

Files in This Skill

cardiology-visual-system/
├── SKILL.md                    # This file
├── scripts/
│   ├── fal_image.py           # Fal.ai image generation
│   ├── gemini_infographic.py  # Gemini infographic generation
│   ├── plotly_charts.py       # Common chart templates
│   ├── lida_quick_viz.py      # LIDA prototyping (⚠️ prototype only)
│   └── convert_slides.sh      # Marp conversion helper
├── templates/
│   ├── marp_medical.md        # Medical slide template
│   └── plotly_medical.py      # Medical chart templates
└── references/
    └── prompt_examples.md     # Curated prompts for each tool