Stone Burden and Density: Mapping Kidney Stones in 3D for Endourological Planning

Two questions decide the operation

Kidney stone disease is common and rising: the lifetime prevalence of urolithiasis is estimated at around 10%, with recurrence in roughly half of patients within ten years. When a stone becomes a surgical problem, non-contrast computed tomography (NCCT) is the imaging gold standard — and almost every treatment decision that follows is driven by two numbers read off that scan:

1. How much stone is there? — the stone burden
2. How hard is it? — the stone density

Stone burden and density, together with location and renal anatomy, determine whether a patient is best served by shockwave lithotripsy (SWL), retrograde intrarenal surgery (RIRS) with a flexible ureteroscope and laser, or percutaneous nephrolithotomy (PCNL). The European Association of Urology (EAU) guidelines build their treatment algorithm directly on these parameters.

The problem is that both numbers are routinely simplified on 2D images. This article looks at why, and at how AI-driven 3D segmentation restores the quantitative detail that the axial slice throws away.

Stone burden: why diameter is the wrong unit

In everyday practice, stone size is reported as a single maximum linear diameter — "an 11 mm lower-pole stone." That number anchors the EAU size thresholds that steer treatment:

Renal stone size	Typical first-line options
< 10 mm	SWL or RIRS
10–20 mm	SWL or RIRS (location- and density-dependent)
> 20 mm	PCNL favored

The trouble is that a single diameter is a poor proxy for the actual quantity of stone that has to be fragmented and removed. A stone is a three-dimensional object, and:

• Linear measurement is plane-dependent. The longest axis may not lie in the axial plane, so the reported diameter under- or over-estimates the true maximum. Inter-observer variability in caliper placement is well documented.
• The ellipsoid formula overestimates volume. The common shortcut — estimating volume from three diameters using a sphere or ellipsoid formula — systematically overstates the real volume of irregular stones. Finch and colleagues showed that 3D software reconstruction yields substantially different (and more accurate) volumes than the ellipsoid approximation.
• Burden is cumulative. A patient with multiple stones, or a branched (staghorn) calculus, cannot be captured by one diameter at all. What predicts operative time, stone-free rate, and the need for staged procedures is the total volumetric burden, not the largest fragment.

Volume is the honest unit. Two "15 mm" stones can differ in true volume by a factor of two or more depending on shape. Reporting burden as a segmented volume — per stone, per kidney, and as a total — gives a measure that actually tracks the work of the operation.

Stone density: Hounsfield units predict success

The second number is attenuation, measured in Hounsfield units (HU) on the NCCT. Density matters for two reasons.

It predicts how the stone will fragment

Higher-attenuation stones resist shockwave lithotripsy. Across multiple series, mean attenuation above roughly 1000 HU is associated with significantly lower SWL disintegration and stone-free rates; El-Nahas and colleagues identified stone density on non-contrast CT as an independent predictor of SWL failure. For dense stones, the EAU algorithm leans toward endoscopic approaches (RIRS or PCNL) where laser energy does the fragmentation rather than the shockwave.

It hints at composition

Attenuation correlates — imperfectly — with stone chemistry. As an approximate guide on single-energy CT:

Stone composition	Approximate attenuation
Uric acid	~200–450 HU
Struvite	~600–900 HU
Cystine	~600–1100 HU
Calcium oxalate dihydrate	~1000–1200 HU
Calcium oxalate monohydrate / calcium phosphate	> 1200 HU

These ranges overlap, so a single mean HU value cannot reliably name the mineral — dual-energy CT does that far better. But the directional signal is clinically real: a low-density stone raises the possibility of uric acid (potentially dissolvable with oral chemolysis), while a very high-density homogeneous stone warns of poor shockwave fragmentation.

Heterogeneity matters as much as the mean

A single average HU hides how density is distributed inside the stone. A homogeneous, uniformly dense stone fragments differently from one with a soft core and a hard shell. The internal density distribution — not just the peak — is what the surgeon actually wants to see.

What the 3D density layers show

This is exactly what jst/medics renders for kidney stones. Rather than collapsing a stone to one HU number, the model segments it into cumulative density bands and stacks them as nested, color-graded shells:

Layer	Threshold	Reading
Band 0	0+ HU	Full stone envelope
Band 1	300+ HU	Moderately mineralized volume
Band 2	700+ HU	Dense core
Band 3	1000+ HU	Hardest, shockwave-resistant fraction

Each band is cumulative — every voxel above the threshold — so peeling the layers from outside in reveals the density gradient of the stone in three dimensions. A stone that is uniformly purple at the 1000+ HU band is a poor SWL candidate everywhere; a stone with only a small high-density core may fragment from the outside while the core resists. That distribution is invisible on a single axial slice and unrepresentable in a single HU figure.

How AI turns a CT into a quantitative map

The bridge from "a stack of grayscale slices" to "a segmented, measured 3D object" is automated segmentation. A convolutional neural network labels the stone voxels on the CT, and from that labeling the system derives, without manual caliper work:

• Volumetric stone burden — true segmented volume rather than an ellipsoid estimate, reported per stone
• Per-kidney assignment — each stone allocated to its side and, where relevant, its calyceal group
• Density mapping — the cumulative HU bands above, computed voxel-by-voxel rather than from a single region of interest
• A rotatable model — the spatial relationship between stone, collecting system, and renal anatomy explored directly instead of reconstructed mentally across slices

The same shift that 3D reconstruction brought to renal-tumor nephrometry — replacing mental reconstruction with direct visual exploration — applies to stones. (We covered that transition for renal masses in the RENAL score article.)

Nephrolithometry scores: stones have their RENAL, too

Just as the RENAL and PADUA scores standardize renal-tumor complexity, nephrolithometry scoring systems standardize stone complexity to predict stone-free rates and complications:

• Guy's Stone Score — grades PCNL complexity from stone number, location, and anatomy
• S.T.O.N.E. nephrolithometry — Stone size, Tract length, Obstruction, Number of involved calyces, and Essence (density), scored directly off the NCCT
• CROES nomogram — predicts PCNL stone-free rate from a multivariable model
• R.I.R.S. and Resorlu–Unsal scores — analogous tools for retrograde intrarenal surgery

Every one of these systems is fed by the same raw inputs — stone size/volume, number, location, and density — and every one is easier to compute reliably when those inputs are read off a segmented 3D model rather than estimated by hand on 2D slices.

A practical example

Consider a 16 mm lower-pole stone — squarely in the 10–20 mm band where SWL and RIRS are both on the table. The decision hinges on detail:

1. Burden: segmented volume confirms it is a compact 16 mm stone, not a 16 mm-long thin fragment — a genuine moderate burden.
2. Density: the density bands show a dense core fully filling the 1000+ HU layer — a poor shockwave candidate.
3. Location: the rotatable model confirms a steep, narrow lower-pole infundibulum, where fragments clear poorly after SWL.

Read together, these three facts point away from SWL and toward RIRS with laser lithotripsy — a conclusion that a single line of report text ("16 mm lower-pole stone") leaves wide open. The 3D map does not make the decision; it makes the decision legible.

Conclusions

Treatment selection in stone disease rests on two quantities — how much stone and how hard — yet both are routinely flattened into a single diameter and a single HU value on 2D imaging. Volumetric burden corrects the first; cumulative density mapping corrects the second.

AI-driven 3D segmentation does not replace the urologist's judgment or the EAU algorithm. It supplies those algorithms with better inputs: an honest volume instead of an ellipsoid estimate, a density gradient instead of an average, and a rotatable map instead of a mental reconstruction. For the surgeon, that means a more confident choice between SWL, RIRS, and PCNL. For the patient, fewer surprises and fewer staged procedures.

References

1. Skolarikos A, Neisius A, Petřík A, et al. EAU Guidelines on Urolithiasis. European Association of Urology, 2024. uroweb.org

2. El-Nahas AR, El-Assmy AM, Mansour O, Sheir KZ. A prospective multivariate analysis of factors predicting stone disintegration by extracorporeal shock wave lithotripsy: the value of high-resolution noncontrast computed tomography. Eur Urol. 2007 Jun;51(6):1688-1693. PubMed

3. Mostafavi MR, Ernst RD, Saltzman B. Accurate determination of chemical composition of urinary calculi by spiral computerized tomography. J Urol. 1998 Mar;159(3):673-675. PubMed

4. Finch W, Johnston R, Shaida N, Winterbottom A, Wiseman O. Measuring stone volume - three-dimensional software reconstruction and an algorithm versus the linear measurement and ellipsoid formula. BJU Int. 2014 Apr;113(4):610-614. PubMed

5. Okhunov Z, Friedlander JI, George AK, et al. S.T.O.N.E. nephrolithometry: novel surgical classification system for kidney calculi. Urology. 2013 Jun;81(6):1154-1159. PubMed

6. Thomas K, Smith NC, Hegarty N, Glass JM. The Guy's stone score—grading the complexity of percutaneous nephrolithotomy procedures. Urology. 2011 Aug;78(2):277-281. PubMed