A two-dimensional section shows where a cutting plane intersects a three-dimensional structure. It does not show the whole structure, and its visible profiles are not miniature versions of complete objects. A small circular profile might belong to a small particle, or it might be a shallow cut through a much larger one.
Stereology connects measurements on sections to quantities such as volume, surface area, length and number through geometric probability and a defined sampling design. These quantities require different measurement rules; counting visible profiles cannot answer every question. The ATS/ERS standards for quantitative assessment of lung structure make this distinction explicit for measurements derived from microscopic sections.
The practical task is to separate what an image shows directly from what a study can legitimately estimate. That distinction determines whether a result describes the original specimen or just the appearance of selected slices.
What Happens When a Three-Dimensional Object Is Sectioned?
Consider a sphere with a radius of 10 micrometers. A plane passing through its center produces a circular profile with a diameter of 20 micrometers. A plane passing 8 micrometers from the center produces a profile with a diameter of only 12 micrometers. Both profiles come from exactly the same sphere.
The geometry follows the relationship:
r2 = R2 − z2
Here, r is the profile radius, R is the sphere radius, and z is the distance between the section plane and the sphere’s center. Without knowing the cutting position, the profile diameter alone does not identify the sphere diameter.
Orientation introduces another ambiguity. A straight circular cylinder produces a circular profile when cut perpendicular to its axis, an ellipse when cut obliquely, and an elongated profile when cut along its length. A curved tube can intersect one plane several times, producing separate profiles that belong to one connected structure.
These examples explain why recognizing a shape is different from measuring its three-dimensional properties. A circular outline does not prove that the original object was spherical. A long outline does not, by itself, prove that the object was longer than another object with a round outline.
Geometric assumptions can support an estimate when they are justified, but they must remain visible in the reasoning. The distinction between assumptions about objects and randomization in the measurement process is covered in design-based versus model-based stereology.
Why Profiles Do Not Represent Objects Equally
Larger objects have more opportunities to be intersected
In an ideal population of spheres sampled by randomly positioned, negligibly thin planes, a sphere with twice the diameter spans twice the range of plane positions that would intersect it. It therefore has twice the opportunity to appear in a section, assuming comparable placement within the sampling region and ignoring boundary effects.
This creates two different issues. A large sphere can produce a small profile when cut near its edge, yet large spheres are also more likely to enter the profile sample. Measuring many profile diameters does not automatically make their distribution match the distribution of particle diameters.
More profiles do not necessarily mean more cells
Suppose a hypothetical treatment enlarges cells without changing their total number or the volume containing them. The enlarged cells can intersect more section planes. An increase in profiles per unit section area could then be mistaken for an increase in cell number.
The disector addresses number estimation by using a three-dimensional sampling rule rather than counting every profile present in one plane. Its original formulation uses paired sections with known separation to identify counting events independent of particle size and shape, subject to the method’s sampling and identification requirements. See Sterio’s original paper on unbiased particle counting with the disector.
The distinction is also visible in the units. Profiles per square millimeter describe an areal density. Cells per cubic millimeter describe a numerical density. Total cell number describes a count within a defined region. Changing the label does not change the measurement.
Which Three-Dimensional Quantities Can Sections Estimate?
The measurement should follow the question, not whichever image-analysis button is easiest to press. Volume, surface area, length and number describe different properties, so they require different links between the specimen and the sampling probe.
| Target quantity | Suitable measurement approach | Misleading shortcut |
|---|---|---|
| Volume fraction | Area or point fractions in appropriately sampled sections | Treating one selected field as representative of the specimen |
| Total volume | Areas across sections with known spacing | Extrapolating from one central cross-section |
| Surface area | Boundary intersections with probes under a suitable orientation design | Reporting profile perimeter as three-dimensional surface area |
| Length | Intersections with suitable planes or spatial probes | Adding visible profile lengths in an arbitrary section |
| Particle number | Disector counting with a defined sampling scheme | Equating separate profiles with separate objects |
Area fraction can estimate volume fraction
A component occupying part of a specimen’s volume also occupies part of appropriately sampled section areas. This relationship permits volume fraction estimation without assuming that the component consists of spheres, cylinders or any other ideal shape.
Point counting provides a practical implementation. Count test points falling on the component and divide by the points falling within its reference region. The resulting point fraction estimates the component’s volume fraction when the sampling and classification rules are appropriate. The procedure is developed in the guide to point counting for volume fraction estimation.
For a hypothetical example, suppose 120 of 600 eligible points hit a mineral phase across a properly sampled set of sections. The estimated volume fraction is 20%. That does not mean 20% of the particles belong to that phase, nor does it establish their mean size. A fraction of space is not a fraction of objects.
Pooling also deserves attention. If fields contain different amounts of the reference region, taking an unweighted average of their percentages can answer a different question from dividing the total component points by the total eligible reference points.
More detailed tracing is not always the best use of effort
Measuring every pixel in a few images does not remove variation between sampled regions. Research comparing point counting and image-analysis methods demonstrated that variation between sections can dominate the final sampling error. The practical implication is to balance measurement effort within images against coverage of the specimen, rather than pursuing extremely precise measurements of too few fields. This tradeoff was examined in research on measurement error and sampling variation in stereology.
Section Position and Section Orientation Are Different Problems
Position determines which parts of the specimen enter the sample. Orientation determines the directions from which structures are intersected. Randomizing one does not necessarily randomize the other.
Consider a block containing parallel fibers. Sections taken at evenly spread locations but all perpendicular to the fibers will show mostly transverse profiles. Sections parallel to the fibers will show long profiles. Both sets may cover the block spatially, yet their measurements respond differently to the fibers’ direction.
Surface estimation can accommodate directional structure through an appropriate design. Isotropic sampling gives directions equal representation. Alternatively, vertical sections retain a chosen axis and use a compatible test system; this is not equivalent to applying ordinary straight-line probes to any convenient longitudinal section. The mathematical basis is established in Baddeley, Gundersen and Cruz-Orive’s paper on surface estimation from vertical sections.
Volume estimation does not generally require isotropic section orientation. Parallel sections can support a valid volume estimate if their positions and measurements follow the appropriate design. “Random sections” is therefore too vague for a methods description: readers need to know what was randomized and why.
How Sampled Sections Can Represent a Whole Specimen
A practical design often starts with a random position and continues at a fixed interval through the specimen. This spreads observations across its extent rather than concentrating them in visually interesting regions.
For example, if every tenth section is required, select the starting section randomly from the first ten and retain every tenth section thereafter. Selecting section five because it looks satisfactory, then following a regular interval, is not the same procedure. The guide to systematic uniform random sampling covers its implementation across sections and microscope fields.
Sampling decisions also need to remain consistent with the target. A study of an entire organ cannot simply omit an inconvenient region and retain the original whole-organ interpretation. If a compartment is deliberately excluded, define the target accordingly.
Volume from a series of areas
The Cavalieri approach makes the connection between sections and volume particularly clear. With an appropriate random start and parallel sections at a known constant interval, volume is estimated by multiplying the sum of section areas by that interval. This approach and its sampling efficiency are treated in Gundersen and Jensen’s research on systematic sampling in stereology.
Suppose a hypothetical series has a summed area of 48 square millimeters, and the sampled planes are 0.5 millimeters apart. The estimated volume is:
48 mm2 × 0.5 mm = 24 mm3
The spacing is the distance between sampled planes, not automatically the thickness of an individual section. If intermediate sections were skipped, multiplying by section thickness alone would use the wrong interval.
Why Density and Total Amount Must Be Kept Separate
A density has a numerator and a denominator. Either can change. Before interpreting a difference, write both out in full: cells per volume of which region, or component volume per volume of which reference space?
Consider two hypothetical specimens, each containing 100,000 cells. Specimen A has a reference volume of 100 cubic millimeters; specimen B has a reference volume of 80 cubic millimeters. Their numerical densities are 1,000 and 1,250 cells per cubic millimeter. Specimen B has a 25% higher density despite containing exactly the same number of cells.
The reverse can also happen. Two specimens can have the same density but different total numbers because their reference volumes differ. This is the reference trap: interpreting a ratio as though it were an absolute amount.
Use the wording of the research question to decide what to report. “What proportion of this tissue is connective tissue?” asks for a fraction. “How much connective tissue is present in the organ?” asks for an absolute volume. Those are related questions, not interchangeable ones.
A Section Is Not a Projection or a Reconstruction
An ideal section represents an intersection at a plane. A projection combines information across depth. A reconstruction attempts to recover spatial relationships from a series of observations. Keep these categories separate when choosing images for measurement.
A simple thought experiment shows why. Place two nonoverlapping objects at different depths but at the same horizontal coordinates. A projection can place their images on top of each other. An isolated plane may intersect one or neither. The resulting pictures answer different geometric questions even though they depict the same specimen.
Stereological estimation does not require a complete reconstruction for every target. Conversely, estimating volume or number does not reveal the exact shape, position or connections of every object. If the question concerns which branch joins which vessel, a global length estimate will not answer it.
Check What Preparation Has Done to the Geometry
The specimen measured on a slide may no longer have its original dimensions. Fixation, embedding, cutting and mounting can alter geometry, so a result needs a clear reference state. Practical checks belong in the study design, including those covered in section thickness, guard zones and tissue shrinkage.
For an illustrative calculation, uniform shrinkage to 90% of the original length along each axis leaves 81% of the original area and 72.9% of the original volume. A 10% linear change is therefore not a 10% volume change. Real deformation may be uneven, so this example is not a universal correction factor.
Before interpreting any section measurement, identify the three-dimensional quantity being estimated, the region it refers to, and the sampling rule connecting the observations to that quantity. A clear image helps identify structures. A valid measurement design is what turns those structures into defensible quantitative evidence.