Pooling X-ray detector regions before taking a logarithm usually gives a different answer from taking each region’s logarithm first and averaging afterwards. The detector adds transmitted signals. A logarithm describes those signals through a nonlinear transformation. Changing their order changes the quantity being calculated.
The short answer is to preserve the question. For the transmission of a combined region, add the corrected sample signals and their matching reference signals, form the ratio, then take its negative natural logarithm. For an average of resolved path attenuations, transform the individual ratios and use the weights appropriate to that average. Neither answer should silently acquire the other’s meaning.
This guide develops the distinction using fictional positive signals and an ideal monoenergetic model. It concerns spatially different paths, rather than the different photon energies examined in the existing material-identification guide. All quantities are supplied learning data. The calculations are not equipment settings, clinical interpretations, inspection acceptance criteria or instructions for using an X-ray source.
eduKateSG · X-ray Science
Find your next learning step
Learn why pooled X-ray transmission and average path attenuation differ, using reference weights, equivalent thickness, original examples and transfer checks.
ROUTE 2 · CHAPTERS 5–8
Check weighting and information loss
Work through changed conditions and original examples.
ROUTE 3 · CHAPTERS 9–12
Separate physical and measurement effects
Preserve the evidence needed for the next conclusion.
ROUTE 4 · CHAPTERS 13–18
Test calculations and transfer
Apply the method independently and check its limits.
ROUTE 5 · CHAPTERS 19–20
Answer questions and continue
Use the FAQs, subject routes and verified sources.
Chapter contents
Build the signal model · Chapters 1–4
Check weighting and information loss · Chapters 5–8
Separate physical and measurement effects · Chapters 9–12
Test calculations and transfer · Chapters 13–18
- 13. Distinguish heterogeneity from counting variation
- 14. Combine groups without losing their original contribution weights
- 15. Recognise a limited approximation for small variation
- 16. Attempt a changed three-region problem
- 17. Build a practical evidence check
- 18. Choose a learning route from the first unstable step
Answer questions and continue · Chapters 19–20
Imagine a detector region with a corrected reference signal of 1,000 units when the object is absent. With the object in the stated path, the corrected signal is 500 units. Its transmission is T = 500/1,000 = 0.5. Transmission here is a dimensionless response ratio, not a displayed shade or an independently established material property.
For an ideal narrow monoenergetic beam through a homogeneous material, write T = exp(−μt). Here μ is a linear attenuation coefficient and t is path thickness, with compatible units. Their product p = μt equals −ln(T). More generally, the ideal line-path quantity can integrate a varying coefficient along the path. This article uses the simpler constant-coefficient case when converting to thickness.
NIST's explanation of the mass attenuation coefficient states the exponential law and distinguishes mass thickness from geometric thickness. Our notation uses geometric t and linear μ. Confusing those conventions would introduce a unit error before any pooling question begins.
First check that the reference belongs to this region and the stated measurement conditions. A ratio made with a different region’s illumination does not become meaningful merely because both numbers carry the same signal unit. For now, treat all supplied values as exact model inputs. Later chapters separate this structural issue from noise, background and detector response.
CHAPTER 2 OF 20 · Build the signal model
2. Understand what the logarithm accomplishes
A logarithm reverses exponentiation. In this model it turns multiplicative transmission along one path into an additive attenuation quantity. If one layer transmits half the incoming signal and a second independent layer on the same path transmits half of what reaches it, the overall transmission is one quarter. The logarithmic values add: −ln(0.25) = 2 × −ln(0.5).
Numerically, −ln(0.5) is approximately 0.6931 and −ln(0.25) is approximately 1.3863. The larger p means a smaller transmitted fraction under this model. It does not mean that a detector directly counted 1.3863 particles or that the object absorbed that numerical amount of energy.
The addition applies to successive contributions along a path. It does not establish that neighbouring paths should be added in logarithmic form to reproduce a detector’s total signal. Neighbouring regions contribute signals side by side; successive layers act on the radiation remaining along the same path. These are different physical arrangements.
A useful first check is to draw where each multiplication happens. Does the radiation encounter both layers successively, or does one part of the beam encounter one path while another part encounters another? That drawing often resolves a mistaken calculation before formal algebra is needed. Keep the relationship between physical arrangement and mathematical operation visible as the examples become less uniform.
Consider two separate regions, A and B. Each has a corrected reference signal of 1,000 units. Their sample signals are 800 and 200 units. The resolved transmissions are therefore 0.8 and 0.2, with logarithmic attenuations approximately 0.2231 and 1.6094.
If the question concerns the combined detector response, the sample total is 1,000 and the reference total is 2,000. The pooled transmission is 0.5 and its negative logarithm is approximately 0.6931. If the question concerns the equally weighted average of the two resolved p values, the result is approximately 0.9163. Both computations are reproducible. They describe different summaries.
| Quantity | Region A | Region B | Combined or averaged result |
|---|---|---|---|
| Corrected reference signal | 1,000 | 1,000 | 2,000 total |
| Corrected sample signal | 800 | 200 | 1,000 total |
| Transmission | 0.8 | 0.2 | 0.5 pooled |
| Negative natural logarithm | 0.2231 | 1.6094 | 0.6931 after pooling |
| Equally weighted resolved p | — | — | 0.9163 |
Pause before deciding which answer is correct. What does the requested result mean? A combined signal measurement calls for one calculation. A mean over resolved paths calls for the other. The mistake would be to report 0.6931 as the mean of the resolved attenuations, or 0.9163 as the negative logarithm of the pooled transmission, without declaring an approximation that changes the question.
CHAPTER 4 OF 20 · Build the signal model
4. Follow the signal weights into the pooled ratio
For region i, let Rᵢ be its corrected reference signal, Sᵢ its corrected sample signal and Tᵢ = Sᵢ/Rᵢ. Adding sample and reference signals gives Tpool = ΣSᵢ/ΣRᵢ. Substituting Sᵢ = RᵢTᵢ shows that Tpool = ΣwᵢTᵢ, where wᵢ = Rᵢ/ΣRᵢ.
Those weights are reference-response fractions. They add to one and describe how strongly each region contributes to the reference total. Equal reference signals give equal weights. Equal geometrical areas alone do not prove equal reference signals, because illumination and the measurement system can differ across the field.
The notation becomes easier if you read it as a contribution account. Each reference contribution is reduced by its own transmission. The reduced contributions are added, then divided by the original total. That is why an arithmetic mean of all regional transmissions is appropriate only when the reference weights are equal, or when equal weighting defines a separate desired quantity.
Write the weights beside the data before calculating a logarithm. This simple habit prevents two errors from becoming entangled: choosing the wrong average and changing the order of a nonlinear transformation. It also makes a later review possible. A final p value alone does not reveal which reference contributions, regions or exclusions created it.
CHAPTER 5 OF 20 · Check weighting and information loss
5. Change the illumination without changing the paths
Now give A a reference signal of 3,000 units and B a reference signal of 1,000 units. Keep their transmissions at 0.8 and 0.2. Their sample signals become 2,400 and 200. The pooled ratio is 2,600/4,000 = 0.65, and −ln(0.65) is approximately 0.4308.
An unweighted average of the transmissions still gives 0.5, but it no longer reproduces the combined detector response. A contributes three quarters of the reference response. The appropriate response-weighted average is 0.75 × 0.8 + 0.25 × 0.2 = 0.65.
If a separate question requests the reference-weighted mean of resolved p, calculate 0.75 × 0.2231 + 0.25 × 1.6094, approximately 0.5697. This remains different from 0.4308. Equal-area mean attenuation would require area weights, which are a third definition if areas and reference responses differ.
This changed case is a useful transfer test. A learner who remembers that the first example used an ordinary average may repeat it automatically. A learner who understands the contribution account reconstructs the weights from the new references. The physical transmissions have stayed fixed while their contribution to the pooled response has changed. Neither a changed pooled value nor an unchanged regional value alone establishes that the material changed.
CHAPTER 6 OF 20 · Check weighting and information loss
6. See what a pooled value cannot recover
Return to equal reference signals. One object arrangement gives regional transmissions 0.8 and 0.2. Another gives 0.5 and 0.5. Both have pooled transmission 0.5. The first has unequal resolved attenuations with mean approximately 0.9163; the second has equal resolved attenuations with mean approximately 0.6931.
The combined signal therefore does not uniquely recover the distribution of path attenuations. Many pairs can share the same sum. A single pooled measurement has deliberately discarded the information needed to distinguish them. More decimal places in that pooled number cannot restore the missing spatial information.
The ambiguity can be larger than the first pair suggests. With equal weights and pooled transmission fixed at 0.5, one transmission can approach zero while the other approaches one. The average negative logarithm can then grow without a finite upper bound in this ideal model. A real measurement has limits, but the mathematical example shows why a pooled ratio alone does not supply a useful upper bound on mean path attenuation.
An informative report names the pooled region and preserves resolved data when available and appropriate. If the fine-scale data were never acquired or retained, say that their variation is unresolved. Do not fill the missing spatial distribution with an assumption of uniformity and then describe that assumption as an observed property of the object.
CHAPTER 7 OF 20 · Check weighting and information loss
7. Explain the direction of the difference
For positive transmissions and nonnegative weights adding to one, the negative logarithm is a convex function. Consequently, −ln(ΣwᵢTᵢ) is no larger than Σwᵢ[−ln(Tᵢ)]. The comparison uses the same weights on both sides. Changing from response weights to area weights removes that particular comparison.
For two equally weighted regions, the same result follows from the arithmetic–geometric mean relationship. Their arithmetic mean transmission is at least their geometric mean. Applying a decreasing negative logarithm reverses that ordering. The negative logarithm of the geometric mean equals the arithmetic mean of their negative logarithms.
In the original example, the arithmetic mean transmission is 0.5 and the geometric mean is √(0.8 × 0.2) = 0.4. Their negative logarithms are approximately 0.6931 and 0.9163. This provides an independent route to checking both the value and the direction of the gap.
Equality occurs when all positively weighted regional transmissions are equal. It can also appear trivially when only one region has nonzero weight. In the ideal positive model, unequal transmissions with positive weights give a strict difference. This is a statement about specified mathematical quantities, not a general instruction that one type of image processing is always superior. The right quantity still depends on the question and the validated measurement model.
CHAPTER 8 OF 20 · Check weighting and information loss
8. Convert to equivalent thickness only under a stated model
Suppose both regions contain the same fictional homogeneous material at one photon energy, with stipulated μ = 0.5 per centimetre. Each resolved thickness is then tᵢ = pᵢ/μ. The 0.8 path gives approximately 0.4463 centimetres; the 0.2 path gives approximately 3.2189 centimetres. Their equally weighted mean thickness is approximately 1.8326 centimetres.
If the pooled transmission 0.5 is instead fitted by one uniform slab of that material, the equivalent thickness is 0.6931/0.5, approximately 1.3863 centimetres. It is the thickness of a uniform model slab reproducing the pooled transmission under these assumptions. It is not the arithmetic mean of the original two thicknesses.
This distinction gives the mathematical difference a physical interpretation. More transmitting paths contribute disproportionately to the remaining signal. A uniform equivalent can therefore be thinner than the corresponding weighted average of heterogeneous paths, even when every path contains the same stipulated material.
The conversion fails to have that simple thickness meaning if material coefficients differ between regions, the spectrum is broad, the coefficient is unknown or the detector model does not fit. A reported equivalent thickness must carry its material, energy, geometry and weighting assumptions. Without them, a number in centimetres can look more physically direct than the underlying evidence permits. The general lesson is to name the equivalence being constructed, rather than let an inverse calculation masquerade as a direct geometrical observation.
CHAPTER 9 OF 20 · Separate physical and measurement effects
9. Keep side-by-side paths separate from stacked layers
Imagine a first layer with transmission 0.8 and a second with transmission 0.2. If every relevant ray crosses both successively, their combined transmission is 0.8 × 0.2 = 0.16 in the ideal model. Its logarithmic attenuation is approximately 1.8326, the sum of the two layer values.
Place those alternatives side by side with equal reference weights instead, so half the response travels through each alternative path. The pooled transmission becomes 0.5 and its logarithmic value approximately 0.6931. The same two transmission numbers have led to different calculations because the arrangement changed.
Do not interpret this as a complete real-object simulation. Scatter, spectrum, geometry and material boundaries can introduce additional conditions. The example isolates a simpler reasoning question: are contributions sequential along each path, or alternative contributions across the measured footprint?
A useful classroom task supplies the two drawings without an equation label. Ask students to decide which needs a product and which needs a weighted sum, then justify the decision in words. Follow by changing the side-by-side weights to one quarter and three quarters. The response should change because the contribution account changes, not because the learner has found another memorised formula. Spatial arrangement belongs inside the model, rather than being treated as decoration beside the arithmetic.
CHAPTER 10 OF 20 · Separate physical and measurement effects
10. Test how a small open fraction affects a strongly attenuating region
Consider an ideal footprint in which 99% of the reference response follows paths with transmission 0.01, while 1% follows open paths with transmission 1. The pooled transmission is 0.99 × 0.01 + 0.01 × 1 = 0.0199. The apparently small open contribution has nearly doubled the pooled transmission relative to 0.01.
The corresponding negative logarithm is approximately 3.9170, compared with 4.6052 for a uniformly 0.01-transmitting footprint. The open fraction is small in reference weight but large relative to the very weak transmitted signal from the other paths. Percentages need a denominator: 1% of the reference response is not 1% of the final response.
Under this exact two-path model, an independently known attenuated transmission a and an observed pooled T would imply open reference fraction f = (T − a)/(1 − a). Substituting a = 0.01 and T = 0.0199 gives f = 0.01. Without independently justified a and the two-path assumptions, this inversion does not establish a real object’s gap fraction.
This is a sensitivity lesson, not a defect-detection threshold. Other mixtures can reproduce the same pooled value. An image feature requires evidence about resolution, geometry, calibration and alternative explanations before anyone assigns a physical cause. The simple example explains why unresolved contributions can matter while leaving actual identification to the appropriate evidence.
CHAPTER 11 OF 20 · Separate physical and measurement effects
11. Separate spatial mixing from spectral mixing
So far, all regional paths use an ideal single photon energy. The difference between pooling and averaging logarithms therefore does not require a broad spectrum. Heterogeneous paths alone are enough. That is the distinct gap this article addresses.
A broad spectrum introduces a different mixture: different photon energies can have different attenuation and detector response. The existing guide to why one transmission reading cannot identify a material develops that ambiguity and includes a two-component spectral example. Spatial and spectral mixtures can coexist, but one should not be offered as proof of the other.
Glover and Pelc’s original research on nonlinear partial-volume artefacts establishes a related imaging principle: integrating spatially varying attenuation before the logarithm can create nonlinearity even with monochromatic radiation. The numerical cases here are independent teaching constructions, rather than a reproduction of their study or an implementation of a correction method.
In a real image, a dark band or changed mean value does not identify its cause by itself. A spatial averaging effect, spectrum, scatter or another part of the acquisition and processing chain may be relevant. Our model demonstrates a possible mechanism under explicit conditions. It does not diagnose an artefact from appearance or prescribe how an actual acquisition should be altered.
CHAPTER 12 OF 20 · Separate physical and measurement effects
12. Preserve corrected signals rather than substitute display brightness
The pooled-ratio calculation assumes signals that can legitimately be added under the stated response model. Displayed image shades may have been transformed, clipped, inverted or mapped through a window. Averaging those displayed values does not automatically reproduce an average transmission or a pooled physical signal.
Suppose a supplied linear-signal exercise reports reference 1,100 and sample 600, including a known additive offset of 100 in each. The corrected ratio is 500/1,000 = 0.5. Dividing the uncorrected values gives approximately 0.5455. Their negative logarithms differ even before regions are pooled. The offset and its justification belong in the record.
The example does not establish that every background should be corrected by subtracting one common number. Different regions or times may require different supported terms. Some processing can produce nonpositive corrected values, for which the real logarithm used here is undefined. Replacing them with an arbitrary tiny positive value changes the result and requires its own justified method.
For an evidence-reading task, ask where each number came from: raw detector output, corrected response, transmission, logarithmic projection, reconstruction or display. A label such as pixel value is too broad to settle that question. Keeping those stages distinct helps students see why a mathematically correct average can still be attached to the wrong physical quantity.
CHAPTER 13 OF 20 · Test calculations and transfer
13. Distinguish heterogeneity from counting variation
The worked regions describe different underlying transmissions. Random fluctuations introduce another reason that measured ratios can differ. If several observations of one uniform path fluctuate, their unequal recorded values do not establish spatial heterogeneity. Conversely, a stable pooled average does not establish uniformity across the footprint.
The logarithm’s nonlinearity also means that transforming noisy measurements can alter their statistical behaviour. Averaging transformed values and transforming an averaged signal need not have identical statistical properties. Zero counts, background estimates, reference uncertainty and detector noise require a suitable model rather than an automatic use of the exact positive-number examples.
For a supplied classroom dataset, separate two questions. What difference follows algebraically from the listed positive values? What evidence supports treating those values as resolved differences in the object? The first can be answered without knowing their uncertainty. The second cannot.
Do not use more averaging as a universal repair. Combining regions may reduce some random variation while discarding spatial detail and changing the estimand, meaning the quantity intended for estimation. Keeping regions separate may preserve detail while leaving larger uncertainty in each. The choice belongs to the actual measurement purpose and validated analysis. A useful learning explanation shows the trade-off in terms of information retained and the question answered, rather than promising that either finer detail or smoother appearance is always more truthful.
CHAPTER 14 OF 20 · Test calculations and transfer
14. Combine groups without losing their original contribution weights
Suppose a dataset has already been divided into groups. For each group g, preserve its summed corrected reference Rg and pooled transmission Tg. Its corrected sample contribution is RgTg. Combining groups then gives Ttotal = ΣRgTg/ΣRg. Group labels alone should not determine the weights.
If only a group’s logarithmic value pg remains, recover its model transmission as exp(−pg) before using that formula. Averaging the group logarithms generally answers another question. Likewise, averaging group transmissions equally is inappropriate when groups represent different reference totals.
Take three groups with reference totals 1,000, 1,000 and 2,000, and transmissions 0.8, 0.2 and 0.5. Their sample totals are 800, 200 and 1,000. The combined transmission is 2,000/4,000 = 0.5. First combining the initial pair gives reference 2,000 and transmission 0.5, so combining that group with the third preserves the result.
This consistency is a useful independent check. Regrouping unchanged signal contributions should not change the final pooled transmission when the correct totals are retained. If regrouping does change it, inspect whether reference weights were dropped or transformed summaries were averaged as though they were raw contributions. The record should make the grouping reversible to the extent needed for the question. Once only one final pooled number survives, the finer spatial account cannot be reconstructed uniquely.
CHAPTER 15 OF 20 · Test calculations and transfer
15. Recognise a limited approximation for small variation
When regional p values differ only slightly, a mean-p approximation may be close to the pooled result. Closeness is a quantitative question, however, rather than permission to assume equality. A useful small-variation expansion is ppool ≈ mean(p) − variance(p)/2, using the same weights and sufficiently small deviations. Higher-order terms matter as variation grows.
For two equal-weight paths with p = 0.9 and 1.1, the mean is 1 and the variance is 0.01. The approximation gives 0.995. Directly calculating −ln[(exp(−0.9) + exp(−1.1))/2] gives approximately 0.995008. In this constructed case, the correction explains both the direction and the small size of the gap.
This approximation comes from expanding the exponential around the weighted mean and then expanding the logarithm. It should not be transferred automatically to strongly different paths, unknown distributions or data distorted by measurement limits. The earlier 0.8 and 0.2 transmission example is better handled directly than described as a tiny perturbation.
For beginners, the exact two-region arithmetic is sufficient. For extension learners, the approximation shows how variation itself enters a pooled logarithmic result. Ask which variance is being used and under which weights. That question keeps an elegant formula connected to the actual collection of paths rather than allowing it to become a generic correction attached to any image average.
CHAPTER 16 OF 20 · Test calculations and transfer
16. Attempt a changed three-region problem
Here is a fresh model. Three regions contribute reference fractions 0.2, 0.3 and 0.5. Their transmissions are 1, 0.5 and 0.1. Find the pooled transmission, its logarithmic value and the reference-weighted mean of resolved logarithmic values. Then identify which region contributes most strongly to the transmitted total.
Work from the contributions first. They are 0.2 × 1 = 0.2, 0.3 × 0.5 = 0.15 and 0.5 × 0.1 = 0.05. The pooled transmission is 0.4 and its negative logarithm approximately 0.9163. The weighted mean of resolved p is 0.2 × 0 + 0.3 × 0.6931 + 0.5 × 2.3026, approximately 1.3592.
The transmitted contributions, divided by their sum 0.4, are 50%, 37.5% and 12.5%. The region supplying only 20% of the reference response supplies half the transmitted response. The region supplying half the reference supplies only one eighth of what remains.
That final comparison checks understanding beyond formula selection. Initial reference weights and final transmitted fractions describe different populations of contributions. If a learner uses final fractions to reconstruct the original pooled transmission, the attenuation has effectively been included twice. Keep the before-and-after accounts separate, and check that each set sums to the correct total before assigning a physical interpretation.
CHAPTER 17 OF 20 · Test calculations and transfer
17. Build a practical evidence check
An interpretable pooled result needs the region definition, reference and sample relationship, signal corrections, weighting and order of operations. It should also identify whether the desired quantity is pooled transmission, its logarithm, an average path attenuation or a model-equivalent thickness. Those are useful distinctions to place beside the result, rather than hide in a distant methods paragraph.
Check that every included region has a matching positive reference under the model. Account for masked regions and changing footprints. If an exclusion changes the represented object or illumination, a comparison may change even when the remaining regional transmissions do not. A difference between two summaries then needs both data and scope examined.
For an independent arithmetic check, reconstruct the combined sample signal from reference contributions and individual transmissions. Confirm that the pooled transmission lies between the smallest and largest included positive transmissions. With the same nonnegative weights, confirm that its negative logarithm is no greater than the weighted mean of regional negative logarithms.
These checks can expose errors, but passing them does not validate the apparatus or identify the material. They establish internal consistency within the supplied model. A dependable explanation carries both statements: what the calculations check and which physical or professional questions remain outside them. That is how a compact numerical result remains useful when passed to another reader.
A further check concerns signal scaling. Multiply every corrected reference and sample signal by the same positive factor, such as five. The pooled transmission and all regional ratios stay unchanged. This is an invariance of the ratio model, so a changed answer would expose an arithmetic or matching error.
Scaling only one region’s reference and sample contributions leaves its individual transmission unchanged but changes its weight in the summed response. That is exactly why a pooled detector-response result should not be described automatically as an area-weighted object property. The signal representation and the physical question need to remain connected. A region-specific gain correction, for example, requires a justified response model before signals are treated as comparable additive contributions. The invariance test checks the bookkeeping; it does not establish that a real detector’s calibration or correction is valid.
CHAPTER 18 OF 20 · Test calculations and transfer
18. Choose a learning route from the first unstable step
If ratios are uncertain, begin with one reference and one sample. Ask what 500/1,000 means and why its unit cancels. Keep logarithms aside until the learner can reconstruct a sample signal from its reference and transmission. Then introduce the logarithm as a transformation of a known positive ratio.
For stabilisation, use the two-region table. Require two labelled answers and a sentence explaining the difference. Change the illumination while holding transmissions fixed. A student who can explain why the pooled answer changes has moved beyond copying the first calculation’s equal weights.
For extension, use the three-region transfer, the small-variation approximation or the non-uniqueness example. Ask the learner to construct a second spatial arrangement with the same pooled transmission. This makes information loss visible through a counterexample they can defend independently.
A useful final task omits topic labels. Present one stacked-layer drawing, one side-by-side drawing and one already-grouped signal record. The learner selects a product, a weighted sum or a contribution reconstruction and explains why. The Education parent guide develops this broader goal of learning that remains usable when the surface details change. Progress appears in choosing and defending the operation, rather than reproducing a familiar answer after being told which formula to use.
CHAPTER 19 OF 20 · Answer questions and continue
19. Questions about pooled attenuation
Is taking the logarithm after pooling wrong?
It is correct for the logarithm of the pooled transmission under the stated additive signal model. It becomes misleading when labelled as a different quantity, such as the arithmetic mean of resolved path attenuations. Define the desired summary before selecting the calculation.
Does equal detector area guarantee equal weights?
Not for a pooled signal ratio. Its weights come from corrected reference contributions. Equal areas can have unequal illumination or response. An area-weighted average is a separately defined quantity and needs its own justified weights.
Can monochromatic radiation still show this nonlinearity?
Yes. Spatially varying paths and integration before a logarithm suffice in the ideal example. Spectral mixing is another issue. Observing one mechanism’s mathematical possibility does not establish the cause of a particular real image feature.
Does a pooled value identify average thickness?
Only a specifically defined model-equivalent thickness follows when the material coefficient and other assumptions are supplied. It generally differs from the corresponding mean of resolved thicknesses. Without the model, the transmission alone does not identify thickness uniquely.
Can a zero signal be logged?
The real logarithm here requires a positive argument. A recorded zero or nonpositive corrected value needs an appropriate measurement-limit and statistical treatment. Replacing it with an arbitrary positive constant changes the result and does not recover an exact attenuation.
What should I retain for checking later?
Keep the appropriate original and corrected signals, matching references, regions, exclusions, transformation order and uncertainty information. A screenshot or final pooled number usually cannot reproduce that evidence chain. Preserve enough information for the intended review and question.
CHAPTER 20 OF 20 · Answer questions and continue
20. Continue through the connected science routes
Understand transmission and its limits
Return to The Gold Standard of X-ray Science for the measurement chain. Use Why One Transmission Reading Cannot Identify a Material for non-unique inverse problems, layers and spectral mixtures. These guides answer adjacent questions without turning a pooled signal into a material identity.
Connect projections with reconstructed images
The existing X-ray Computed Tomography guide connects projections, reconstruction, display and partial-volume limitations. Its image-level discussion complements the signal-order calculations here. Broader subject discovery belongs in How X Works.
Check the scientific foundations
NIST's attenuation-law explanation supports the single-path model and unit conventions. Glover and Pelc's 1980 research abstract supports the distinction between spatial integration and logarithmic attenuation in nonlinear partial-volume effects. These sources establish the identified principles; all tables, teaching cases and transfer calculations in this guide are original hypothetical examples. Source routes checked on 22 October 2026.