Skip to content

imaging.md

Latest commit

 

History

History
223 lines (153 loc) · 6.58 KB

imaging.md

File metadata and controls

223 lines (153 loc) · 6.58 KB
Raw

Imaging Analysis Prompts

Cell segmentation, quantification, and phenotyping for H&E and multiplexed immunofluorescence (MxIF) images.

Servers used: mcp-openimagedata (5 tools), mcp-deepcell (3 tools: segment_cells, quantify_markers, generate_segmentation_overlay), mcp-cell-classify (3 tools: classify_cell_states, classify_multi_marker, generate_phenotype_visualization)

Quick Start: Test Data

GCS Location: gs://sample-inputs-patientone/mcp-deepcell-test-data/test_data/

Available Test Images:

  • dapi_512x512.tif - Nuclear stain (DAPI), ~30 cells
  • dapi_1024x1024.tif - Nuclear stain, ~120 cells
  • dapi_2048x2048.tif - Nuclear stain, ~480 cells (large image test)
  • ki67_512x512.tif - Proliferation marker (Ki67)
  • tp53_512x512.tif - Tumor suppressor marker (TP53)
  • membrane_512x512.tif - Membrane stain

All images are 16-bit TIFF, synthetic data with known characteristics.

Cell Segmentation

Nuclear Segmentation (Basic)

Segment cells from DAPI nuclear staining:

Segment cells in the test image:
- Image: gs://sample-inputs-patientone/mcp-deepcell-test-data/test_data/dapi_512x512.tif
- Use nuclear segmentation model
- Minimum cell size: 100 pixels

Expected: ~30 cells detected with unique IDs and quality metrics.

Membrane Segmentation

Segment whole cells from membrane staining:

Segment cells using membrane marker:
- Image: gs://sample-inputs-patientone/mcp-deepcell-test-data/test_data/membrane_512x512.tif
- Model: membrane (Mesmer)
- Minimum cell size: 150 pixels

Expected: ~30 cells with larger masks (whole cell vs nuclear only).

Large Image Segmentation

Process large images with automatic tiling:

Segment the large test image:
- Image: gs://sample-inputs-patientone/mcp-deepcell-test-data/test_data/dapi_2048x2048.tif
- Model: nuclear
- Minimum cell size: 200 pixels
- Tile size: 512x512

Expected: ~480 cells, automatic tiling, longer processing time (~30-60s first run).

Cell State Classification (mcp-cell-classify)

Single Marker Phenotyping

Classify proliferating vs quiescent cells using Ki67 (requires segmentation mask from mcp-deepcell):

Classify cell states using Ki67 marker:
- Nuclear image: gs://sample-inputs-patientone/mcp-deepcell-test-data/test_data/dapi_512x512.tif
- Marker image: gs://sample-inputs-patientone/mcp-deepcell-test-data/test_data/ki67_512x512.tif
- Marker name: Ki67
- Intensity threshold: 5000

Expected: ~30% Ki67+ (proliferating), ~70% Ki67- (quiescent).

Multi-Marker Phenotyping

Classify cells with multiple markers (uses mcp-deepcell for segmentation, mcp-cell-classify for classification):

Perform multi-marker cell phenotyping:
1. Segment cells from: gs://sample-inputs-patientone/mcp-deepcell-test-data/test_data/dapi_1024x1024.tif
2. Classify with Ki67: gs://sample-inputs-patientone/mcp-deepcell-test-data/test_data/ki67_1024x1024.tif (threshold: 5000)
3. Classify with TP53: gs://sample-inputs-patientone/mcp-deepcell-test-data/test_data/tp53_1024x1024.tif (threshold: 4000)
4. Generate phenotype visualization

Expected: ~120 cells with 4 phenotypes (Ki67+/TP53+, Ki67+/TP53-, Ki67-/TP53+, Ki67-/TP53-).

MxIF (Multiplexed Immunofluorescence) Workflow

Complete MxIF Analysis

End-to-end segmentation (mcp-deepcell) and phenotyping (mcp-cell-classify):

For PatientOne MxIF data:
1. Load composite MxIF image (DAPI + Ki67 + TP53 channels)
2. Segment cells using nuclear model (DAPI channel)
3. Classify proliferation status (Ki67 threshold: 5000)
4. Classify TP53 status (TP53 threshold: 4000)
5. Generate segmentation overlay showing cell boundaries
6. Generate phenotype visualization with color-coded markers
7. Report cell counts per phenotype and spatial distribution

Expected: Cell masks, phenotype classifications, visualizations, spatial statistics.

Visualization

Segmentation Overlay

Create boundary overlays on original images:

Generate segmentation overlay:
- Original image: gs://sample-inputs-patientone/mcp-deepcell-test-data/test_data/dapi_512x512.tif
- Segmentation mask: [from previous segmentation]
- Boundary color: red
- Boundary thickness: 2 pixels

Expected: Original image with cell boundaries highlighted.

Phenotype Visualization (mcp-cell-classify)

Multi-marker phenotype visualization:

Generate phenotype visualization:
- Segmentation mask: [from previous segmentation]
- Phenotype classifications: [from multi-marker classification]
- Color scheme: Ki67+ (green), TP53+ (red), Both (yellow), Neither (blue)

Expected: Color-coded visualization showing phenotype distribution.

Performance Benchmarks

First Run (Cold Start):

  • Model download: ~30-45s
  • 512x512 segmentation: ~5-10s
  • Total: ~35-55s

Subsequent Runs (Warm):

  • 512x512: ~2-5s
  • 1024x1024: ~8-15s
  • 2048x2048: ~30-60s (with tiling)

Best Practices

Image Requirements

  • Format: 16-bit TIFF (grayscale)
  • Size: Any (auto-tiling for >2048x2048)
  • Staining: Nuclear (DAPI, Hoechst) or Membrane (pan-cytokeratin, CD45)

Parameter Selection

  • Nuclear min_cell_size: 50-200 pixels (depends on magnification)
  • Membrane min_cell_size: 100-300 pixels (whole cells larger than nuclei)
  • Intensity threshold: Test on control samples to calibrate

Quality Control

  1. Verify cell count matches manual count (±10%)
  2. Check segmentation mask visually for over/under-segmentation
  3. Validate marker thresholds on positive/negative controls
  4. Compare phenotype percentages to known markers

Common Variables

Image Paths

  • Test data: gs://sample-inputs-patientone/mcp-deepcell-test-data/test_data/
  • PatientOne: gs://sample-inputs-patientone/PAT001-OVC-2025/imaging/

Model Types

  • nuclear - Nuclear segmentation (DAPI, Hoechst, etc.)
  • membrane - Whole cell segmentation (Mesmer model)

Typical Thresholds

  • Ki67 (proliferation): 3000-8000 (calibrate per antibody)
  • TP53: 2000-6000
  • CD8 (T cells): 4000-8000
  • Membrane markers: 5000-10000

Related Resources

Last Updated: 2026-02-09 Status: ✅ Production ready (mcp-deepcell: segmentation + quantification, mcp-cell-classify: classification + visualization)