⚡ electrical engineering

Geometric, acquisition and image-formation limits of two-dimensional ultrasound volumetry across simulated fluid-filled structures

This virtual imaging trial demonstrates that fixed-coefficient two-dimensional ultrasound volumetry for fluid-filled structures suffers from inherent geometry-dependent errors and additional acquisition and image-formation limitations, with no single formula proving uniformly optimal across diverse morphologies.

Original authors: Wesley da Silva Oliveira

Published 2026-10-15

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Wesley da Silva Oliveira

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Measuring the size of a fluid-filled space inside the human body is a common task in medicine, often done with ultrasound. This technology uses sound waves to create images, allowing doctors to see organs without surgery. For decades, the standard way to estimate the volume of these spaces, such as a bladder or a cyst, has been to take a few simple measurements of length, width, and depth from a two-dimensional image. These numbers are then multiplied by a fixed mathematical constant, a single number chosen to represent the average shape of that organ. The method is fast and accessible, but it relies on a hidden assumption: that the organ being measured looks roughly like the average shape the formula was designed for. When the actual shape of the organ differs from that average, the calculation can drift far from the true volume, potentially leading to misdiagnosis or incorrect monitoring of a patient's condition.

A researcher at the Instituto de Pesquisas Energéticas e Nucleares in Brazil set out to understand exactly where this method goes wrong. Instead of testing on patients, where many variables are hard to control, the scientist built a massive, fully digital experiment. This virtual trial allowed them to separate the different sources of error. They could look at the error caused simply by the shape of the object, the error caused by the angle at which the measurement was taken, and the error caused by the way the ultrasound image itself is formed. By creating thousands of perfect digital models of fluid-filled cavities and then simulating the process of scanning them, the team could see how much each step contributed to the final mistake.

The study began by generating over two thousand unique, perfectly defined shapes that resembled a human bladder. These were not real bladders but precise geometric models with known volumes ranging from 10 to 500 milliliters. The researchers first calculated the volume using the standard formulas on these perfect models, assuming the measurements were taken from the exact center and at the perfect right angles. Even under these ideal conditions, where no human error or image distortion existed, the formulas were not perfect. The most common formula, which uses a specific constant to estimate volume, missed the true volume by a median of 12.6 percent. This revealed that a significant portion of the error comes from the formula itself, which forces a complex, irregular shape into a simple geometric box.

Next, the researchers introduced the reality of how a scan is actually performed. In a real exam, the sonographer might not find the perfect center of the organ, or the probe might be tilted slightly. The simulation added these small rotations and shifts to the measurements. This step increased the error further. For the most common formula, the median error rose to nearly 14.5 percent, and the number of cases where the error exceeded a critical 15 percent threshold jumped to almost half of all the measurements. Interestingly, the simulation showed that these acquisition errors did not always make the result worse; sometimes, a slightly off-center measurement accidentally compensated for the formula's inherent bias, bringing the result closer to the truth. However, relying on this luck is not a viable strategy for medical care.

The final layer of the experiment added the physics of image formation. Real ultrasound images are not perfect line drawings; they are built from sound waves that scatter and blur, making the edges of organs appear slightly fuzzy. The researchers simulated this process to see how much the image itself distorted the measurements. They found that the image formation shifted the apparent size of the organ by less than a millimeter on average. While this seems small, it still changed the final volume calculation. More importantly, the simulation revealed a distinct failure mode: for the largest bladders in the study, the organ was simply too big to fit inside the single frame of the ultrasound image. In these cases, the measurement could not be taken at all, regardless of how good the formula was. This "field of view" limitation accounted for a separate 2.4 percent of failures, occurring only in the largest volume group.

The study then tested whether these findings held true for other types of organs. The researchers created digital models for simple cysts, gallbladders, and dilated kidney systems, each with their own unique shapes. The results showed that no single formula worked best for every situation. For simple, round cysts, the standard formula performed very well, with errors staying low. However, for gallbladders, which are often elongated and pear-shaped, the error rates skyrocketed, with most measurements failing the accuracy threshold. For the kidney systems, which have a funnel-like shape, the performance of the formulas shifted again, with a different formula performing best. This proved that the "best" formula depends entirely on the specific shape of the organ being measured.

The core finding is that the error in two-dimensional ultrasound volume measurements is not just a matter of poor technique or bad images. A large part of the inaccuracy is built into the method itself, arising from the fact that a single fixed number cannot accurately describe the volume of every possible shape. The study suggests that while these formulas are useful for quick estimates, they should be understood as model-based approximations rather than direct geometric truths. When the shape of the organ changes, or when the scan is not perfectly centered, the error grows in ways that a simple adjustment cannot fix. The research indicates that for situations requiring high precision, especially when the anatomy is irregular or the organ is large, relying on a single two-dimensional measurement with a fixed coefficient may not be sufficient, and alternative approaches may be necessary to capture the true volume.

Technical Summary: Geometric, Acquisition, and Image-Formation Limits of 2D Ultrasound Volumetry

Problem Statement
Two-dimensional (2D) ultrasound volumetry is a standard clinical practice for estimating fluid-filled cavity volumes (e.g., bladder, gallbladder, renal pelvis) using a few orthogonal linear dimensions multiplied by a fixed geometric coefficient. While convenient, this method embeds a fundamental shape assumption: that specific measured dimensions correspond to a specific true volume regardless of morphological variation. The paper identifies that errors in 2D volumetry arise from three distinct, often conflated stages:

Geometric Inaccuracy: The formula itself may be mathematically inaccurate for the specific cavity shape, even if the true maximal dimensions are known.
Acquisition Error: The scanning plane may miss the maximal cross-section due to rotation or offset.
Image-Formation Error: Physics-limited image formation and interface localization may shift the apparent fluid-tissue boundary.

Conventional clinical and phantom studies typically combine these effects, making it difficult to isolate the contribution of each. This study aims to disentangle these error sources and determine if a single fixed coefficient can be universally optimal across different cavity morphologies.

Methodology
The author conducted a fully computational virtual imaging trial using a nested, three-layer design to control geometry, acquisition, and imaging physics independently.

Study Design: The trial utilized synthetic geometries and predefined acquisition planes. The design space was "frozen" prior to outcome generation to prevent retrospective bias.
Layer 1: Exact Geometry (G): 2,048 bladder-like geometries were generated using a scrambled Sobol low-discrepancy design. These included variations in volume (10–500 mL), aspect ratios, superelliptic shape exponents, pearness, and asymmetry. No pathology (stones, masses) was included. This layer measured intrinsic formula error using exact orthogonal planes and known dimensions.
Layer 2: Acquisition Perturbation (A): Each geometry underwent 12 acquisitions (1 ideal + 11 perturbed). Perturbations included independent plane rotations (-20° to +20°) and offsets (-0.20 to +0.20 of cavity radius). This layer isolated the error introduced by non-ideal plane selection without image formation.
Layer 3: Synthetic Ultrasound (U): Using the PyMUST framework (based on SIMUS), synthetic B-mode images were generated for a stratified subset of 256 acquisitions. This layer added physics-limited image formation, scatterer simulation, and a target-aware localizer to identify the fluid-tissue interface. Five acquisitions were prospectively classified as "unimageable" due to single-frame field-of-view (FOV) limitations.
Formulas Evaluated: Three conventional formulas were tested:
- $V = 0.52 \times L \times W \times D_T$
- $V = 0.2\pi \times L \times W \times D_T$
- $V = 0.625 \times L \times W \times \frac{D_L + D_T}{2}$
Transferability Extension: To test generalizability, the study prospectively added three additional geometry families: simple-cyst-like (triaxial ellipsoids), gallbladder-like (elongated, pear-shaped), and dilated renal collecting-system-like (funnel-shaped).
Metrics: The primary metric was Absolute Percentage Error (APE), with a prespecified failure threshold of >15%.

Key Results

Intrinsic Geometric Error: Significant error existed even in the "perfect" G layer. For the bladder-like design, the median APE for the 0.52 formula was 12.57% (38.9% failure rate >15%), while 0.2π and 0.625 performed better (7.74% and 7.75% median APE, respectively). This confirms that fixed coefficients cannot perfectly model variable morphologies.
Acquisition Impact: Perturbing the acquisition plane (Layer A) generally increased error, though in some cases, it partially compensated for pre-existing geometric bias. The median APE for the 0.52 formula rose to 14.49%, with a failure rate of 47.5%.
Image Formation and FOV: The synthetic ultrasound layer (U) introduced submillimetric mean dimensional shifts (e.g., -0.59 mm for length). While small, these shifts altered volumetric error. Crucially, a distinct FOV limitation mechanism was identified: 2.39% of bladder-like acquisitions were unimageable in a single frame, occurring exclusively in the largest volume quartile (>188 mL).
Morphology-Dependent Performance: No single formula was universally optimal.
- Simple Cyst-like: The 0.52 formula was nearly exact (median APE 0.72% in G; 6.26% in U) and outperformed others significantly, consistent with the theoretical coefficient for a triaxial ellipsoid ( $\pi/6 \approx 0.524$ ).
- Gallbladder-like: All formulas performed poorly due to taper and pearness. The 0.52 formula had a median APE of 26.35% in U, with a 62.9% failure rate.
- Renal Collecting System: The 0.52 formula remained the most accurate (8.18% median APE, 27.2% failure rate), while the 0.625 formula showed the highest failure rate (57.7%).
Effective Coefficient ( $k_{true}$ ): The study calculated the effective coefficient ( $k_{true} = V_{true} / (L \times W \times D_T)$ ). This value varied significantly based on shape exponents (Spearman $\rho=0.825$ with longitudinal shape exponent) rather than volume, demonstrating that morphology, not size, drives the deviation from fixed coefficients.

Significance and Claims
The paper claims that the limitations of 2D ultrasound volumetry are not solely due to image quality or operator error but are fundamentally geometry-dependent.

Separation of Error Sources: The study successfully isolates that a substantial portion of volumetric error arises from the mismatch between the fixed geometric coefficient and the actual cavity morphology, existing even before acquisition or image formation occurs.
No Universal Formula: The results demonstrate that no single fixed coefficient is optimal across different organ morphologies. A formula that works well for simple cysts (0.52) fails significantly for gallbladder-like shapes, and vice versa.
Limitations of Repeatability: High measurement repeatability does not guarantee agreement with the true volume. As seen in the gallbladder extension, technically acceptable images with precise interface localization can still yield high volumetric errors if the underlying shape model (ellipsoid) is inappropriate.
Field of View as a Failure Mode: Single-frame 2D volumetry has a hard limit defined by the probe's field of view. Large targets may be unimageable regardless of the formula used, representing a failure mechanism distinct from calculation error.

Conclusion
The author concludes that 2D ultrasound volumetry should be interpreted as a model-based measurement rather than a direct geometric truth. Its validity is strictly conditional on the morphology of the target and the acquisition geometry. When high accuracy is required, reliance on a single fixed coefficient is insufficient; solutions may require morphology-aware calibration, protocols that ensure true maximal plane acquisition, or a shift to three-dimensional measurements when anatomy or field-of-view constraints render ellipsoid assumptions unreliable.

Technical Summary: Geometric, Acquisition, and Image-Formation Limits of 2D Ultrasound Volumetry

More like this