ALEVA

Insights

Ultrasound Contrast Metrics: Separate Pixel Ordering from Display Remapping

How can an ultrasound OEM distinguish contrast changes caused by reconstruction from changes caused by display remapping? A source-led review of gCNR, the 2026 Contrast Order paper and a reproducible synthetic ordering check.

By ALEVA — Orange County, California, USA

When a new ultrasound beamformer produces a cleaner-looking target, what evidence shows that the improvement survives a change in display mapping? This question matters when an OEM team compares reconstruction software whose outputs use different amplitude scales. A contrast result needs both a defined quantity and a reproducible estimator.

Start with the quantity being measured

Conventional contrast-to-noise ratio uses regional summary statistics. Rodriguez-Molares and colleagues showed that changing dynamic range can improve this metric without a corresponding improvement in pixel separation. Their generalized contrast-to-noise ratio, gCNR, instead uses the overlap between two probability distributions: gCNR = 1 − ∫ min[p(x), q(x)] dx. This is a distribution-separation measure, not a complete assessment of an image or a clinical outcome. [1]

The distinction between the population quantity and its numerical estimate is essential. Schlunk and Byram demonstrated that histogram choices can make estimated gCNR sensitive to sample availability and extreme transformations. They investigated alternatives including rank-based histograms, variable-width bins and empirical cumulative distributions. Their paper was published in 2023; it is a methodological foundation, not a new 2026 result. [2]

A recent ordering-based alternative

Hyun, Beuret and Dahl's Contrast Order paper appeared online in IEEE Transactions on Ultrasonics on October 22, 2026. It defines contrast order through the expected sign of the brightness difference between regions. With a fixed convention in which X denotes the target and Y the reference, write CO = P(X > Y) − P(X < Y). Ties contribute zero. This retains the direction of contrast while avoiding a density-estimation step. The authors also introduce an effective contrast ratio calibrated to ideal Rayleigh-speckle statistics. That calibration assumption should accompany any reported ECR value. [3]

The journal record updates a February 2, 2026 arXiv preprint. The journal article and the earlier preprint are separate versions; the preprint is not independent corroborating evidence. The protocol below is an engineering proposal, not a reproduction of the authors' experiments. [3, 4]

A small reproducible ordering check

Consider deliberately synthetic intensity samples X = [1, 2] and Y = [2, 4]. Across the four equally weighted pairs, X is smaller three times, tied once and larger zero times. The empirical contrast order is therefore −3/4. Applying the same strictly increasing transformation f(z) = z² to these positive values produces X′ = [1, 4] and Y′ = [4, 16]. All four ordering outcomes remain unchanged, so the result is still −3/4.

Now clip every value to a maximum of 2. The samples become [1, 2] and [2, 2]. Two pairs are smaller and two are tied, giving −1/2. This is an arithmetic counterexample: clipping is not strictly increasing and can destroy ordering information. Quantization can also introduce ties. A decreasing transformation reverses the sign convention rather than preserving the signed value. These examples are calculations on invented arrays, not measurements of an ALEVA device.

Proposed OEM comparison record

Freeze the analysis contract before evaluating candidate software. Keep the same source acquisition, physical regions and region labels. Record the processing stage at which values are sampled: envelope output, compressed image, or exported display pixels. Save the masks and the treatment of zeros, invalid values and ties. If gCNR is included, specify the estimator and its settings rather than recording only its name.

Use two distinct checks. First, hold the reconstructed image fixed and apply a shared, strictly increasing remapping with adequate numerical precision. This tests the metric implementation's intended invariance. Second, evaluate the actual delivery path, including its clipping and quantization. Record whether ordering changes enter during reconstruction or during display preparation. A failure of the second check should not be silently attributed to the beamformer.

Retain complementary spatial measurements. A regional distribution or ordering statistic does not encode where pixels lie within a region: permuting their positions leaves those statistics unchanged. Consequently, a favorable score alone cannot demonstrate preserved boundaries, point-target response or fine structure. Use application-specific spatial and temporal checks alongside contrast analysis, and report repeat-acquisition variation separately from differences caused by estimator settings.

The handover decision

For an OEM or ODM evaluation, deliver the source identifiers, analysis masks, processing-stage definition, estimator version and remapping checks with the metric values. Set acceptance criteria for the intended imaging task before selecting a winner. Contrast order provides a useful question about relative brightness; gCNR provides a different question about distribution separation. Neither authorizes a claim of improved diagnosis or a complete product-performance conclusion by itself.

Sources and publication status

  1. Rodriguez-Molares A. et al. The Generalized Contrast-to-Noise Ratio: A Formal Definition for Lesion Detectability. IEEE TUFFC, 67(4), 745–759, April 2020; online November 29, 2019. Peer-reviewed. DOI: 10.1109/TUFFC.2019.2956855; public record.

  2. Schlunk S., Byram B. Methods for Enhancing the Robustness of the Generalized Contrast-to-Noise Ratio. IEEE TUFFC, 70(8), 831–842, August 2023; online June 26, 2023. Peer-reviewed. DOI: 10.1109/TUFFC.2023.3289157; university-hosted paper.

  3. Hyun D., Beuret S., Dahl J. J. The Contrast Order: An Order-Based Image Quality Criterion for Nonlinear Beamformers. IEEE Transactions on Ultrasonics, online ahead of print, October 22, 2026. Peer-reviewed journal article; publication metadata and abstract checked. DOI: 10.1109/TUSON.2026.3746169; public record.

  4. Hyun D. The Contrast Order: An Order-Based Image Quality Criterion for Nonlinear Beamformers. arXiv:2602.01622v1, February 2, 2026. Earlier preprint; superseded by the journal record above for publication status.