bioimage-measurement-qc

measure segmented bioimage objects, generate qc summaries, and export tables, plots, scripts, and notebooks for microscopy analysis. use when labels or masks already exist or when a segmentation result needs quantitative validation. prefer this skill for regionprops-style measurements, channel intensity summaries, object filtering, csv export, quality checks, and concise interpretation of likely segmentation or measurement issues.

Bioimage Measurement QC

Overview

Measure labeled objects, export analysis tables, and generate quality-control outputs that make segmentation results easier to trust and debug.

Default stack

Prefer:

numpy
scipy
scikit-image
tifffile
pandas
matplotlib
napari

Inputs

Typical inputs are:

raw image
processed image
labels or binary mask
optional channel metadata

If labels are missing, stop and redirect to the appropriate segmentation workflow unless the user explicitly wants a measurement-from-mask-only task.

Measurement workflow

Confirm which label image is the source of truth.
Confirm whether measurements are 2D, slice-wise 3D, or full 3D.
Measure geometry and intensity separately but report them together.
Export a flat table with one row per object whenever possible.

Default measurements

Prefer a practical default set:

label id
area or volume
centroid
bbox
equivalent diameter
eccentricity when meaningful
solidity when meaningful
mean intensity
max intensity
min intensity

For multichannel data, use explicit column naming such as:

mean_intensity_ch0
mean_intensity_ch1

QC outputs

Generate QC artifacts by default:

histogram of object size
histogram of mean intensity
object count summary
overlay image or napari layer suggestion
summary of likely failure modes

Example failure modes to flag:

many tiny objects -> threshold too low or noise not removed
many border-touching objects -> crop/field-of-view issue
implausibly large merged objects -> under-segmentation
too many zero-intensity objects -> label/image mismatch

Filtering rules

When filtering objects, preserve provenance.

Save both unfiltered and filtered tables when practical.
Record thresholds used for filtering.
Never silently drop rows.

Deliverables

Produce these by default unless the user opts out:

CSV measurement table
QC plots in PNG format
runnable Python script
notebook version
concise narrative summary

Interpretation rules

Do not over-interpret biological meaning from morphology alone.
Separate measurement facts from biological hypotheses.
Flag suspicious units when pixel size metadata is missing.

Resources

references/qc-checklist.md: common QC checks and failure modes
scripts/measure_labels.py: minimal measurement script