The Cavalieri principle estimates the volume of an object from the areas of parallel sections through it. Measure or estimate those areas, add them together, and multiply by the distance between sampled planes. The method does not require the object to resemble a sphere, cylinder or any other convenient shape.
In stereology, the practical version combines this geometry with systematic sampling and a random start. It can estimate the volume of an organ, anatomical region or material component without reconstructing every detail in three dimensions. The statistical basis for this approach is developed in Gundersen and Jensen’s study of systematic sampling and Cavalieri estimation.
The Cavalieri Volume Formula
For equally spaced sampled planes, the estimator is:
V̂ = T × ΣAi
Here, V̂ is estimated volume, T is the perpendicular distance between sampled planes, and Ai is the area of the target structure on sampled plane i. The sum includes all sampled planes intersecting the structure across its full extent.
If areas are measured in mm2 and spacing in mm, the result is in mm3. Keep these units consistent before calculating; a spreadsheet will happily multiply incompatible units without raising an eyebrow.
The estimator is unbiased under its sampling and measurement conditions. This means its average over possible random starts equals the true volume, not that every individual estimate is exact.
For volume alone, the cutting direction does not need to be isotropic, meaning equally likely in every spatial direction. Parallel sections in a fixed anatomical orientation can be used. Orientation can affect precision, however, and additional measurements may impose different requirements. This distinction is established in research on ordinary and isotropic Cavalieri designs.
Choose the Section Interval and Random Start
The random start determines the position of the whole systematic series. For a continuous sampling interval of T, select an offset uniformly within one interval, then place subsequent planes T apart. Continue the sequence through the complete target.
For an existing serial section collection, a common implementation is to select every kth section after randomly choosing the first section from the first k positions. If every tenth section will be analysed, each of the first ten positions must have an equal chance of being selected.
Suppose the random choice is section 7. The sample then contains sections 7, 17, 27, 37 and so on. Choosing section 7 because it looks particularly clear is not random selection, even if every subsequent section follows the correct interval.
The systematic uniform random sampling guide covers the broader sampling design. For a Cavalieri study, record the random start, section sequence and interval before measuring areas. Keep the sequence intact rather than replacing inconvenient sections informally.
Also define the sampling extent before selecting the start. If tissue has already been trimmed away, ask whether part of the target was removed. A correctly sampled remainder does not recover an unrecorded missing end.
Estimate Section Areas with Point Counting
Area measurement does not always require tracing every boundary. With point counting, a calibrated grid is placed over the section and the observer counts test points falling inside the target region.
Each point represents a known area, written as a/p. The estimated area on a section is:
Âi = (a/p) × Pi
Substituting this into the volume equation gives:
V̂ = T × (a/p) × ΣPi
For a square grid with point spacing of 0.20 mm at specimen scale, each point represents 0.04 mm2. Count the designated test point, such as the centre of a cross, rather than any part of the marker touching tissue. Use a predefined rule for points on ambiguous boundaries.
Randomise the grid position within its repeating unit rather than moving it until the pattern appears convenient. Record counts separately for each section, even though the final volume formula uses their sum. Practical use of calibrated grids on image sections is documented in research applying Cavalieri estimation to CT and MR images.
For a new protocol, write down exactly what counts as target tissue: whether cavities are included, how tears are handled, and which anatomical landmarks define the boundary. The companion guide to point counting and volume fractions covers the method beyond its use in Cavalieri estimation.
Worked Example: Calculating a Regional Volume
Consider a hypothetical specimen sampled at intervals of 0.50 mm. Eight sampled sections intersect the target. A point grid represents 0.04 mm2 per point.
| Sampled section | Points inside target | Estimated area, mm2 |
|---|---|---|
| 1 | 12 | 0.48 |
| 2 | 28 | 1.12 |
| 3 | 45 | 1.80 |
| 4 | 56 | 2.24 |
| 5 | 53 | 2.12 |
| 6 | 39 | 1.56 |
| 7 | 22 | 0.88 |
| 8 | 5 | 0.20 |
| Total | 260 | 10.40 |
The estimated volume is:
V̂ = 0.50 × 0.04 × 260 = 5.20 mm3
The same result follows from multiplying the summed estimated areas, 10.40 mm2, by 0.50 mm.
Do not divide by the number of sampled sections. The formula requires the sum of their areas, not their mean. If starting with a mean area, the calculation must also include the number of sampled sections.
Check a second issue before accepting the result: does 0.50 mm describe the distance between sampled planes, or only the thickness of an individual section? Substituting one for the other can change the answer by the entire sampling factor.
How Many Sections and Points Are Enough?
Plan this through a pilot study rather than adopting a fixed section count for every specimen. Retain the ordered section measurements so that precision can be assessed using a method appropriate to systematic sampling.
The coefficient of error, or CE, expresses the sampling uncertainty of an estimate relative to its size. It is not the same as biological variation between subjects. Nor does a low CE prove that boundaries, calibration or processing corrections are correct. Research comparing Cavalieri variance and CE prediction methods shows why the choice of error estimator needs attention.
Use the pilot to compare practical alternatives. Would closer section spacing improve the estimate more than a denser point grid? Are some target profiles so small that the proposed grid rarely hits them? Does the area sequence change smoothly, or does it rise and fall abruptly?
A useful pilot record includes measurement time as well as counts and estimated precision. Compare candidate protocols before committing the full specimen collection. State the precision criterion in advance and retain the evidence used to choose the final design.
Avoid reporting a CE calculated from the grand total of points alone when the chosen method requires section order or other inputs. Two specimens can have the same total point count but very different distributions across their sampled sections.
Section Spacing, Thickness and Tissue Shrinkage
Three measurements are easy to confuse: the instrument’s cutting advance, the final thickness of a mounted section, and the distance between the planes used to estimate volume.
If a serial collection is cut at an actual advance of 50 µm and every tenth section is selected, the sampled planes were nominally 500 µm apart at cutting. That arithmetic does not establish whether 500 µm is appropriate for areas measured after further processing.
The area and distance measurements must refer to compatible physical states. Combining areas measured after shrinkage with an earlier spacing can produce a quantity that represents neither the original nor the final specimen volume. The problem is addressed directly in research on block advance errors in Cavalieri volume estimation.
Start the protocol by stating which volume is wanted: fresh, fixed, embedded or finally processed tissue volume. Then document how each measurement relates to that state. Do not assume that replacing one thickness value with another automatically corrects all deformation.
For illustration, if a specimen retained exactly 90% of its original dimensions in each of three directions, it would retain 0.93, or 72.9%, of its original volume. Real tissue need not shrink uniformly, so this arithmetic is not a universal correction factor.
Keep processing records alongside the sampling data. The guide to section thickness and tissue shrinkage covers how to investigate these measurement mismatches.
Missing Sections and Unequal Intervals
The simple T × ΣA formula assumes a common interval. If a selected section is lost, deleting it and leaving everything else unchanged no longer represents the planned sample. Substituting its neighbour also changes the geometry and should not be treated as an invisible repair.
Keep the missing position in the record. Distinguish between a plane where the target is genuinely absent, which has zero target area, and a plane whose area could not be measured. Missing data are not zero area.
Unequal spacing requires an estimator and error assessment suited to the available section positions and sampling design. Research on Cavalieri estimation from non-equidistant sections develops a trapezoidal estimator for this purpose. It is not a blanket justification for multiplying the remaining area sum by an average interval.
Decide how damaged or missing sections will be handled during protocol development. A documented recovery rule is preferable to a different improvised fix for every specimen.
Using Cavalieri Estimation on Digital Images
Digital tracing and segmentation can replace manual point counting as ways to obtain section areas. They do not remove the need to check what the image boundary represents.
Image slices have finite thickness, and their appearance can differ from an ideal mathematical plane. Partial volume and projection effects can affect estimated boundaries. A primary study of slice thickness in cone beam CT Cavalieri measurements demonstrates why acquisition settings need validation for the object being measured.
For an imaging protocol, record pixel dimensions, plane spacing and slice thickness separately. Check calibration after resizing or exporting images. When using automated masks, inspect representative boundaries before accepting a large batch of calculated volumes.
For validation, consider a suitable phantom or an independent volume measurement, with its own uncertainty documented. Agreement between repeated segmentations tests repeatability; agreement with a valid reference addresses a different question.
What to Report in a Cavalieri Study
A reproducible methods section should let another researcher identify the target, reproduce the sampling sequence and recalculate the volume. Include:
- The target boundary definition and specimen processing state.
- The section orientation, random start procedure and sampling interval.
- The area measurement method and spatial calibration.
- The area per point and section-level counts when using point counting.
- The treatment of missing sections, deformation and thickness measurements.
- The volume estimator, units and method used to assess precision.
Preserve the ordered area or point-count data rather than only the final volume. These records make it possible to check arithmetic, investigate unusual sections and reassess precision. The guide to reporting stereological methods and results covers the wider study record.
The Cavalieri calculation is short. Most of the work lies in defining the target, sampling it fairly and keeping area and spacing measurements physically consistent. Settle those decisions before collecting the main dataset, and the final multiplication becomes the straightforward part it should be.