Measurement Metrics and Their Biological Meaning#

PhenoTypic’s measurement operations extract quantitative features from detected colonies. This page explains what each metric means biologically and when to use it.

For how to apply each metric — which to report directly vs. use as a classifier feature — see Measurement Classification: Phenotypes vs. Features.

Size Metrics (MeasureSize)#

Changed in version 0.20.0: Size magnitudes moved from MeasureShape to MeasureSize, and the radius columns were rebuilt. Size_MedianRadius, Size_MeanRadius and Size_MaxRadius are not the retired Shape_* columns of the same name; see the rename table on the SIZE reference.

Metric

Unit

Biological meaning

Area

pixels

Colony size; proxy for biomass or growth

IntegratedIntensity

sum of pixel values

Total pigmentation; proxy for metabolic output

Perimeter

pixels

Colony boundary length. Increases faster than area for irregular shapes.

ConvexArea

pixels

Area of the colony’s convex hull — its “filled-in” footprint if indentations were removed; a low Solidity (Area/ConvexArea) signals invasive or spreading growth.

BboxArea

pixels

Area of the smallest bounding rectangle around the colony; a low Extent (Area/BboxArea) indicates a spread-out or irregular colony, or possible interference with a neighboring well.

MajorAxisLength / MinorAxisLength

pixels

Fitted ellipse dimensions. The ratio indicates elongation.

InscribedRadius

pixels

Radius of the largest circle that fits entirely inside the colony. Reflects the colony’s narrowest dimension, not its overall extent — an elongated colony reports half its width regardless of length (a 100×20 px colony reports 10), and a runner or spur leaves it unchanged.

MedianRadius

pixels

Median distance from the colony’s center to its boundary, sampled uniformly by angle around every piece of the colony; for an ideal disk it equals the disk radius.

MeanRadius

pixels

Mean distance from the colony’s center to its boundary, sampled by angle; a runner or spur pulls it upward in proportion to its angular width (use RobustMeanRadius for the compact body).

RobustMeanRadius

pixels

Trimmed mean (20% from each end by default) of the same angularly-sampled boundary distances, discounting a runner or spur to estimate the radius of the colony’s compact body.

MaxRadius

pixels

Largest angularly-sampled boundary distance from the colony’s center — its farthest reach; a value far above RobustMeanRadius signals a protrusion, spur, or runner.

Area is the most commonly used growth metric. For calibrated images, convert pixels to physical units (mm²) using the known pixel pitch.

Shape Metrics (MeasureShape)#

Metric

Range

Biological meaning

Circularity

0–1

How round the colony is (1 = perfect circle). Irregular shapes suggest stress, mutation, or sectoring.

Solidity

≈0–1 (can slightly exceed 1)

Ratio of area to convex hull area. Low solidity indicates lobed or branching morphology.

Eccentricity

0–1

Elongation (0 = circular, ~1 = very elongated). Elevated in filamentous or swarming colonies.

Compactness

≥1

Perimeter² / (4π × Area). Equals 1 for a perfect circle; increases with irregularity.

MeanBoundaryDist

pixels

Mean distance from each colony pixel to the nearest edge, computed on the object in isolation — a measure of interior thickness, not a radius. For an ideal disk it is 1/3 of InscribedRadius. Thin appendages on a compact body (runners, a hyphal fringe) pull the ratio below 1/3; a uniformly thin, elongated colony raises it toward 1/2.

MedianBoundaryDist

pixels

Median distance from each colony pixel to the nearest edge — the same interior-thickness measure as MeanBoundaryDist, but more robust to boundary raggedness.

Shape metrics are particularly useful for distinguishing wild-type from mutant morphology, or for detecting contamination.

Intensity Metrics (MeasureIntensity)#

Metric

Biological meaning

MeanIntensity

Average colony brightness; correlates with pigment concentration

MedianIntensity

Robust center of the intensity distribution; less affected by bright/dark spots

StandardDeviationIntensity

Internal color variation; high values suggest heterogeneous pigmentation or sectoring

MinimumIntensity / MaximumIntensity

Intensity extremes within the colony

CoefficientVarianceIntensity

Normalized variation (StdDev / Mean); comparable across colonies of different brightness

Color Metrics (MeasureColor)#

Color measurements report compact, outlier-resistant colorimetric summaries rather than a per-channel statistical suite. The default output covers two perceptually meaningful spaces, CIE L*a*b* and HSV.

Two robust center colors (CIE L*a*b*). L*a*b* is designed so Euclidean distance approximates perceived color difference (ΔE76).

  • L*GeoMedian, a*GeoMedian, b*GeoMedian — the ΔE76 (Euclidean) geometric median of the colony’s pixels. The multivariate median (breakdown point 0.5), so specular highlights, agar bleed-through, and debris cannot drag the center color.

  • L*Medoid, a*Medoid, b*Medoid — the ΔE2000 medoid: the single real colony pixel minimizing total ΔE2000 to all other pixels. Because ΔE2000 is not a true metric, a medoid (which needs no metric axioms) is used instead of a geometric median, and the result is always an actual colony color.

Within-colony color consistency (ΔE2000, from the medoid). A perceptually-corrected uniformity profile:

  • DeltaE2000MedianFromMedoid — robust “perceptual MAD.”

  • DeltaE2000MeanFromMedoid — the color-science uniformity standard.

  • DeltaE2000P95FromMedoid — near-worst-case deviation; flags sectoring and contamination without being dominated by a single stray pixel.

  • LabTotalVariance — the trace of the L*a*b* covariance (var L* + var a* + var b*), a single Euclidean spread scalar.

Robust HSV. Because hue is circular and HSV is not perceptually uniform, each pixel is embedded into Cartesian cone coordinates before the robust center is computed:

  • HueRobustMean, SaturationRobustMean, ValueRobustMean — the cone-embedded geometric-median center converted back to H, S, V (circular-correct; unreliable hue at low saturation collapses toward the achromatic axis automatically).

  • HSVConeVariance — the trace of the cone-Cartesian covariance.

Plot swatch. MedoidColorHex is the sRGB hex string of the ΔE2000 medoid color, intended for visualization only — it is never used as a numeric measurement or analysis input.

These metrics are essential for:

  • Quantifying pigmentation differences between strains

  • Detecting sectoring (regions of different color within a colony)

  • Normalizing color across imaging sessions (after ColorCorrector)

CIE XYZ and xy chromaticity remain available as opt-in legacy per-channel suites (MeasureColor(include_XYZ=True), MeasureColor(include_xy=True)) but are hidden from the default colorimetric output.

Texture Metrics (MeasureTexture)#

Haralick texture features computed from gray-level co-occurrence matrices at specified scales. These capture spatial patterns within colonies that size and shape metrics miss:

  • Smooth vs. rough colony surfaces

  • Concentric ring patterns

  • Internal structure differences between strains

Choosing Metrics#

Biological question

Metrics

Growth rate comparison

Area (time series)

Strain fitness ranking

Area + MeanIntensity

Morphology screening

Circularity + Solidity + Eccentricity

Pigmentation assay

MeanIntensity + CIELAB color channels

Colony heterogeneity

StandardDeviationIntensity + texture features