Property interpretation

Overview

Properties are numerical features extracted from images and/or masks. They quantify morphology, intensity, texture, and image quality. Use them for:

Distance computation (source='properties')
Dimensionality reduction (PCA, t-SNE, UMAP colored by properties)
Correlation analysis with embeddings
Statistical comparison between experimental groups

Properties are computed on the largest connected component of the mask when a mask is used. Multi-channel images produce per-channel properties (e.g., intensity_mean_ch0, intensity_mean_ch1). Pass channel_names=["DAPI", "GFP"] to find_files to get names like intensity_mean_DAPI instead.

Shape Properties (from mask)

These require a binary mask. They describe object size and shape. Computed via skimage.measure.regionprops.

Property	Measures	Range / Formula	Interpretation
area	Foreground pixel count	≥ 0	Larger = bigger object
perimeter	Boundary length (4-connected contour)	≥ 0	Rough edges → higher perimeter
major_axis_length	Major axis of fitted ellipse	≥ 0	Main elongation direction
minor_axis_length	Minor axis of fitted ellipse	≥ 0	Width perpendicular to major
eccentricity	Elongation of fitted ellipse	0 (circle) – 1 (line)	How elongated vs round
solidity	Area / convex hull area	0 – 1	Low = concave/indented shape
orientation	Angle of major axis vs horizontal	[−π/2, π/2] rad	Cell alignment direction
extent	Area / bounding box area	0 – 1	How much of box is filled
equivalent_diameter	Diameter of equal-area circle	2√(area/π)	Size in interpretable units
euler_number	Objects − holes	integer	1 = solid, 0 = has a hole
convex_hull_area	Convex hull pixel count	≥ area	Used internally for solidity

Derived properties (computed from the above):

Property	Formula	Interpretation
circularity	4π · area / perimeter²	1 = perfect circle, < 1 = irregular boundary
aspect_ratio	major_axis / minor_axis	1 = round, > 1 = elongated
roundness	4 · area / (π · major²)	1 = circle, < 1 = elongated

Learn more: scikit-image regionprops reference

Intensity Properties

Whole-image (no mask needed)

Computed over all pixels. Useful for illumination/background characterization.

Property	Measures	Interpretation
intensity_mean	Average pixel intensity	Overall brightness
intensity_std	Standard deviation	Spread / contrast
intensity_min	Minimum intensity	Darkest pixel (baseline)
intensity_max	Maximum intensity	Brightest pixel (saturation)

Masked (inside object only)

Computed only on foreground pixels (mask > 0.5). Useful for fluorescence/staining quantification.

Property	Measures	Interpretation
intensity_mean_masked	Mean inside object	Signal level in segmented region
intensity_std_masked	Std inside object	Intracellular heterogeneity
intensity_min_masked	Min inside object	Darkest point in object
intensity_max_masked	Max inside object	Peak signal

Learn more: NumPy statistical functions

Concentric Ring Properties

The object is divided into concentric rings by Euclidean distance from the boundary (via scipy.ndimage.distance_transform_edt). Ring 1 = outermost (membrane), higher = closer to center (core).

Property	Measures	Interpretation
ring_N_mean	Mean intensity in ring N	Radial intensity profile
ring_N_std	Std of intensity in ring N	Local heterogeneity per ring

Captures spatial gradients — e.g. whether a marker is concentrated at the membrane (ring 1) vs the nucleus (ring N). Useful for translocation assays and protein localisation.

Use create_concentric_ring_function(num_rings=3, stats=['mean','std']) → ring_1_mean, ring_2_mean, ring_3_mean, etc.

Texture Properties (GLCM)

Gray-Level Co-occurrence Matrix features computed within the masked region using skimage.feature.graycomatrix at 4 angles, distance=1, averaged.

Property	Measures	Range	Interpretation
texture_contrast	Squared intensity differences between neighbours	[0, (levels−1)²]	High = grainy/rough
texture_dissimilarity	Absolute intensity differences	[0, levels−1]	High = heterogeneous
texture_homogeneity	Closeness to GLCM diagonal	[0, 1]	High = smooth/uniform
texture_energy	Sum of squared GLCM elements	[0, 1]	High = repetitive/ordered
texture_correlation	Linear dependency between pixel pairs	[−1, 1]	High = predictable pattern
texture_asm	Angular Second Moment (= energy²)	[0, 1]	Orderliness measure

Learn more: scikit-image GLCM tutorial · graycoprops reference

Image Quality Properties

Property	Measures	Range	Interpretation
sharpness_metric	Focus quality (multi-scale blur estimation)	0 (sharp) – 1 (blurry)	QC: filter out-of-focus images
intensity_entropy	Shannon entropy of intensity histogram	≥ 0	Low = uniform; high = complex texture

Learn more: skimage.measure.blur_effect · skimage.measure.shannon_entropy

Choosing properties

Avoid redundancy: circularity and compactness are inverses — use circularity. aspect_ratio and roundness both capture elongation; pick one unless both add value.
Match your question: Morphology → area, perimeter, eccentricity, solidity. Intensity → masked mean/std. Texture → GLCM. QC → sharpness_metric.
Presets: For which features are included in property_preset="basic" / "standard" / "complete" (and other valid presets), see Select properties — Presets.

Topic	Document
Preset bundles	Select properties — Presets
Writing your own functions	Select properties
Extension points	Extending PhenoMe
API reference	`compute_properties`
Concepts overview	Metadata and properties
Workflow examples	Common workflows