Stereology provides a way to estimate three-dimensional quantities from sampled sections or images. It connects what can be measured under a microscope with questions about whole structures: how much tissue is present, how many cells occupy a region, or how much surface separates two phases.
The foundation is not a particular microscope or counting program. It is the relationship between the quantity being estimated, the sampling procedure and the measurement rule. A sharply focused image can still produce a misleading answer if any of those three are mismatched.
These fundamentals support the methods introduced in our broader stereology overview. Before choosing a technique, establish what the result should represent, which parts of the specimen can enter the sample and what observations will support the estimate.
What a Section Can and Cannot Tell You
A section shows intersections between a cutting plane and structures occupying space. Those intersections are profiles, not necessarily complete objects. One cell may appear in several consecutive sections, while another falls entirely between the sections selected for examination.
Consider a spherical particle. A cut through its center produces a large circular profile; a cut near its edge produces a small one. Two circles of different sizes could therefore come from different particles, or from different cutting positions through identical particles. Profile diameter alone does not settle the question.
Orientation creates another ambiguity. A tube cut across its axis looks roughly circular, while the same tube cut along its axis appears elongated. Counting every visible shape as an independent object confuses the appearance of a section with the organization of the specimen.
Stereology addresses these problems through defined sampling and measurement procedures rather than by guessing the missing geometry. The companion guide to three-dimensional structures in two-dimensional sections examines these interpretation problems in more detail.
Define the Quantity Before Choosing the Method
“Measure the tissue” is not a sufficient study objective. Volume, surface area, length and number describe different properties. A treatment could change cell size without changing cell number, or alter the surface of a structure without changing its volume.
The measurement procedure must match the intended quantity. Geometric probes are the points, lines, planes or sampling volumes used to make observations. Their purpose is to turn encounters with the specimen into measurements with a known mathematical relationship to the target.
| Target quantity | Example question | Typical stereological approach |
|---|---|---|
| Volume | How much space does a region occupy? | Section areas combined with known section spacing |
| Volume fraction | What proportion of a specimen is a particular phase? | Points hitting the phase relative to points hitting the reference space |
| Surface area | How much membrane or interface is present? | Intersections with test lines under suitable orientation conditions |
| Length | How much fiber or vessel length is present? | Intersections with appropriately oriented test surfaces |
| Number | How many discrete particles occupy the region? | Counting events within disectors, with an appropriate sampling design |
Point and intersection counting provide practical routes to volume fractions and surface measurements without tracing every boundary. The original paper on practical stereological methods for morphometric cytology develops these relationships and their measurement errors.
Use units as a quick check on the proposed answer. Profiles per square millimeter are not cells per cubic millimeter. Surface area per unit volume is not total surface area. A calculation can be tidy and still describe the wrong quantity.
The Reference Space Determines What the Result Means
The reference space is the region to which a measurement applies. It might be an entire organ, a named anatomical compartment, a biopsy or a manufactured specimen. Define its boundaries before selecting fields or counting structures.
A density expresses an amount relative to that reference space. Numerical density is number per unit volume; surface density is surface area per unit volume. Volume fraction expresses the volume of a component relative to the containing volume. The denominator belongs in the interpretation, not just in a spreadsheet heading.
A hypothetical density comparison
Suppose specimen A contains 1,000 cells in a reference volume of 10 mm3. Its numerical density is 100 cells/mm3. Specimen B also contains 1,000 cells, but its reference volume is 5 mm3. Its density is 200 cells/mm3.
The density has doubled. The number of cells has not changed.
This arithmetic illustrates the reference trap: interpreting a change in a ratio as a change in its numerator. If the question concerns total cell number, density alone leaves the answer unresolved. A valid total-number estimator, or compatible estimates of numerical density and reference volume, is needed.
Write objectives accordingly. “Estimate the volume fraction of collagen within the sampled tissue compartment” is a different task from “estimate total collagen volume in the organ.” Neither result is inherently better; they answer different questions.
Sampling Makes the Estimate Defensible
Sampling determines which parts of a specimen can influence the result. Choosing fields because they look typical gives the observer control over inclusion. A probability sampling procedure replaces that judgment with a reproducible rule.
In a uniform design, eligible locations have equal inclusion probabilities. Other valid designs can use unequal probabilities, provided those probabilities are known and handled correctly in the estimator. The basic requirement is a defensible connection between the sampled material and the intended reference space.
Systematic uniform random sampling combines a random start with regular spacing. For a hypothetical sequence of 100 sections, selecting every tenth section begins by drawing a start from positions 1 through 10. If the start is 7, the selected positions are 7, 17, 27 and so on through 97.
The random start is not decorative. Always starting at section 1 would be a different procedure. The research on systematic sampling efficiency in stereology establishes why distributed sampling can use measurement effort efficiently.
Keep the hierarchy visible: specimens, blocks, sections and fields are different sampling levels. Twenty fields from one specimen do not become twenty independent specimens. Our guide to sampling in stereology covers the selection procedures beyond this introductory framework.
Design-Based and Model-Based Reasoning
Design-based stereology places the basis for inference in the randomized sampling procedure. The specimen is treated as fixed, and the estimator’s behavior is assessed over the samples that the design could produce.
Model-based stereology instead uses a statistical model of the structure. Depending on the approach, that model may describe spatial distributions, orientations or particle geometry. Its suitability depends on whether its assumptions are appropriate for the material and the question.
This distinction is not equivalent to “good method versus bad method.” A justified model can be useful. Trouble starts when an assumption is treated as an observed property, such as assuming spherical particles because their profiles look round.
Nor does design-based mean that laboratory conditions cease to matter. The intended sampling rules must still be followed, and the target structures must remain identifiable. Compare the approaches in design-based versus model-based stereology before deciding what assumptions a study can support.
Why Particle Counting Needs a Third Dimension
Simple profile counts are affected by the opportunity for an object to intersect a section. Objects with greater extent perpendicular to the plane have more opportunities to appear. A change in particle size can therefore change profile counts without changing particle number.
The disector introduces a sampling volume and a counting event defined through depth. A physical disector uses paired sections. An optical disector uses focal planes within a sufficiently thick section. Both require clear rules for identifying objects and deciding which events qualify.
The original disector paper on unbiased particle estimation established a three-dimensional counting approach that does not require a particle-size or particle-shape correction.
The distinction matters when writing an analysis plan. “Count stained profiles” and “estimate the number of stained cells” are not interchangeable instructions. Also define the counting unit: if nuclei are counted, the relationship between nuclei and cells must support any claim about cell number.
Position and Orientation Are Separate Sampling Decisions
A section can have a randomly selected position but a fixed orientation. Those are separate properties. Spreading samples across a specimen does not, by itself, provide the directional sampling needed for every estimator.
Volume estimation can use parallel sections in a fixed direction when the positional sampling is appropriate. Many surface and length estimators need suitable directional randomization because aligned structures interact differently with probes pointing in different directions.
Isotropic uniform random sections provide a design without a preferred direction. Vertical section methods retain a chosen axis while randomizing rotation around it. For surface estimation, they use matching test systems, such as correctly aligned cycloid curves, rather than treating the sections as fully isotropic.
The research on surface area estimation from vertical sections demonstrates how section orientation and probe geometry work together. An arbitrary section cannot be made suitable for every measurement simply by placing a grid over it.
Bias and Precision Answer Different Questions
Bias concerns systematic departure from the target quantity. Precision concerns variation between estimates. An unbiased estimator is correct on average over repeated sampling under its design; an individual estimate need not equal the true value.
A coefficient of error, or CE, expresses the relative sampling uncertainty of an estimate. It does not measure biological variation between specimens, and a small CE does not prove that the sampling or identification rules were correct.
Increasing measurement effort can improve precision, but cannot repair systematic selection errors. This distinction is emphasized in the ATS/ERS standards for quantitative assessment of lung structure, which also address preparation and measurement controls.
Consider two hypothetical plans. One counts many events in a few handpicked fields. The other distributes observations through the defined reference space using a probability design. The first plan may generate a reassuringly large count, but that count does not answer whether omitted regions differ from the selected fields.
For study planning, ask where another hour of work would help: more specimens, more sections or more observations within existing fields. Avoid treating a counting target as a substitute for that decision.
Preparation and Reporting Complete the Measurement
The specimen reaching the microscope may not retain its original dimensions. Fixation, embedding and sectioning can cause shrinkage or distortion. Record the preparation state to which dimensional results apply, and check that reference-volume measurements and microscopic measurements are compatible.
Resolution and labeling also affect what can be recognized. A sound sampling plan cannot recover structures that the imaging method fails to reveal. Record calibration, identification criteria and handling of ambiguous boundaries before analysis begins.
A useful final check is to complete one sentence: “We estimated this quantity, within this reference space, using this sampling procedure and this measurement rule.” If any part remains vague, resolve it before collecting more images.
Keep the supporting record equally concrete: specimen numbers, sampling intervals, measurement units, preparation details and the method used to assess uncertainty. The stereology glossary provides a shared vocabulary for those descriptions. Good stereology makes the path from observation to estimate explicit, so another reader can judge what the result does—and does not—establish.