Feature Extraction
Step 5 – Feature Extraction: turning images into numbers
Section titled “Step 5 – Feature Extraction: turning images into numbers”Computers do not work directly with “tumor” or “stroma” – they work with numbers. Feature extraction is the step where you convert whole-slide images or patches into numeric descriptors, such as color statistics, texture measures, or deep features from a neural network. The result is usually a table where each row is a slide or patch and each column is a feature.
Technical name: Feature Extraction
What this is
Section titled “What this is”Convert images into measurements and numbers:
- Cell counts; nuclear size/shape.
- Gland architecture and textures.
- Region‑level intensity or morphology summaries.
Typical questions
Section titled “Typical questions”- “Can we quantify ‘pleomorphic nuclei’?”
- “How many tumor‑infiltrating lymphocytes are here?”
- “Can I get measurements instead of just a yes/no?”
Common tasks
Section titled “Common tasks”- Segment nuclei/cells.
- Measure areas, perimeters, distances, densities.
- Calculate texture/intensity measures.
- Turn annotations/regions into numeric feature tables.
Core tools (examples)
Section titled “Core tools (examples)”- QuPath — cell detection and intensity/morphology metrics.
- CellProfiler — pipeline‑style cell/tissue measurements.
- scikit‑image — toolbox for custom image features.
- Histolab / TIAToolbox — WSI‑focused feature extraction.
Clinician mental model
Section titled “Clinician mental model”Turn your qualitative description into spreadsheets of measurements—ready for statistics or ML.
Ready-to-use code
Section titled “Ready-to-use code”- FEAT-01: Morphometrics & texture — scikit-image + Pillow blocks for shape (circularity), texture (GLCM), and intensity (hyperchromasia) measurements to turn patches into numeric features.