Figure2 Aaps Agentic Organoid Analysis Prompt

# Figure 2 Concept: Agentic Analysis of Organoid Images with AAPS

Date: 2026-04-29

## User Requirements Captured

The intended Figure 2 should support the paper story:

**Agentic Analysis of organoids with AAPS.**

The core message is that pure agent behavior is difficult to control directly. For biomedical image analysis, the agent needs a structured harness because image quality, imaging conditions, magnification, field content, object scale, and experiment design vary across datasets. The same segmentation or quantification method may not apply to every dataset.

The figure should communicate that AAPS provides an agentic but controlled analysis pipeline with dataset management, image quality control, segmentation method selection, quantification design, database maintenance, plotting, and conclusion generation.

Key required concepts:

1. **Dataset and metadata management**

AAPS should automatically manage datasets from imaging capture, maintain a database, store metadata, and preserve the experiment design.

2. **Quality control and image priors before segmentation**

AAPS should perform quality control and infer image priors for later segmentation. It should choose the proper segmentation strategy based on magnification, image quality, organoid size prior, brightfield/darkfield modality, object distribution, and expected count prior. The pipeline can use a vision model to read and reason about images inside the AAPS workflow.

3. **Flexible segmentation methods**

AAPS should support flexible segmentation methods, including Cellpose, thresholding, multiscale segmentation, and other task-specific methods. For difficult segmentation or specific features that are hard to quantify directly, the pipeline may also use an image generation model to generate segmentation annotations or identify hard-to-quantify visual features.

4. **Agent-designed quantification metrics**

AAPS should use an agent to design quantification metrics based on the experiment design, then run the quantification. Example metric families include growth, differentiation, fusion, morphology, darkness, area, perimeter, centrality, and edge-based structure.

5. **Database, plots, and conclusions**

AAPS should maintain a database of metrics and outputs, generate plots, and produce experiment-level conclusions.

## Figure 2 Design Goal

Create a modern, clear, Nature-style workflow figure that explains why a controlled agent harness is needed for biomedical organoid image analysis and how AAPS converts heterogeneous microscopy data into quality-controlled segmentation, metrics, plots, and conclusions.

The figure should feel like a serious biomedical methods figure, not a marketing diagram.

## Recommended Figure Structure

Suggested panel layout:

- **Panel A: Why uncontrolled agents are risky**
  - Show variable organoid microscopy images entering an unconstrained AI agent cloud.
  - Indicate possible failure modes: inconsistent segmentation, wrong metric choice, poor QC, and weak reproducibility.
  - Keep this compact, as the purpose is to motivate AAPS.

- **Panel B: AAPS as the harness**
  - Show AAPS as a structured control layer around the agent.
  - Include guardrails/checkpoints: metadata, experiment design, QC, method routing, audit trail, database.

- **Panel C: Image-aware routing and segmentation**
  - Show a vision model reading images and extracting priors: magnification, modality, image quality, organoid size, density, and object counts.
  - Show routing to Cellpose, thresholding, multiscale segmentation, or image-generation-assisted annotation.

- **Panel D: Experiment-aware metric design**
  - Show the agent choosing metrics based on biological questions and experiment design.
  - Include growth, differentiation, fusion, darkness, area, perimeter, and centrality/edge metrics as examples.

- **Panel E: Database, plots, and conclusions**
  - Show structured database tables, metric plots, representative overlays, and concise conclusions.
  - Emphasize reproducible outputs and auditability.

## High-Quality Image Generation Prompt

Use this prompt for ChatGPT image generation or another image model: