A seemingly minor choice in ultrasound analysis could substantially change how scientists measure muscle damage after stroke, according to a new study of the calf muscle. Researchers found that the size of the “region of interest”—the area selected within an ultrasound image for computer analysis—strongly influenced the numerical description of muscle tissue. The effect was especially pronounced for two measures, echovariation and mean echointensity, raising concerns that results from different studies may not be directly comparable if researchers select image areas in different ways. The finding turns an apparently technical detail into a potentially important issue for rehabilitation research, where quantitative ultrasound is increasingly being explored as a fast, inexpensive and radiation-free method for tracking changes in muscle structure. The study examined the medial gastrocnemius, the large calf muscle often affected by weakness and abnormal stiffness following stroke, and identified a 300-by-300-pixel sampling area as the most reliable compromise between capturing representative tissue and avoiding surrounding structures.
Ultrasound imaging produces far more information than a conventional clinical scan that is simply viewed by eye. When sound waves pass through muscle, they are reflected differently by contractile fibers, connective tissue, fat and fluid. The resulting pattern of light and dark pixels, known as echotexture, can be quantified using mathematical features. Mean echointensity, or EI, summarizes the average brightness of a selected image region and can reflect changes in tissue composition. Echovariation, or EV, describes the degree to which brightness changes across the region, providing an indication of how uniform or heterogeneous the tissue appears. Other measurements can capture texture at different scales, including the length and uniformity of bright or dark pixel runs and spatial relationships between neighboring gray levels. Together, these variables may reveal changes that are difficult to detect through manual examination alone, including muscle wasting, infiltration by fat or connective tissue, and remodeling associated with spasticity.
The new analysis focused on people living with stroke, whose muscles may undergo a complex combination of weakness, reduced use and altered neural control. Stroke can damage the brain pathways that coordinate movement, often leaving one side of the body less active and causing spasticity—an involuntary increase in muscle tone that can make movement stiff or difficult. The gastrocnemius medialis is particularly relevant because it contributes to pointing the foot downward and stabilizing the leg during standing and walking. Changes in its internal architecture could therefore be related not only to muscle health but also to balance, gait and rehabilitation outcomes. Yet the apparent texture of a muscle on ultrasound depends partly on what portion of the image is analyzed. A large selection may include a broad range of fibers and tissue features, while a small selection may overrepresent a local patch that is unusually bright, dark or structurally irregular.
To test how much this choice mattered, Diego Lapuente-Hernández and colleagues performed a secondary analysis of 130 ultrasound images obtained from 22 post-stroke individuals. From each image, they calculated 21 echotexture features using five different sampling strategies. One strategy covered the maximum visible area of muscle—the whole-muscle region of interest—while the others used square regions measuring 300 by 300, 200 by 200, 100 by 100 and 50 by 50 pixels. The comparison was designed to mimic a practical problem in quantitative imaging: a whole-muscle selection may provide a broad measurement but can be difficult to delineate consistently and may accidentally include skin, fascia or other non-muscular structures. Smaller fixed-size regions are easier to standardize, but they may not represent the muscle as a whole. The researchers used Friedman statistical tests to compare measurements, Bland–Altman analyses to evaluate agreement between approaches, and linear regression models to examine how feature values changed as the selected area became smaller.
The results showed a clear and opposing pattern for the two principal measures. Echovariation decreased significantly as the region of interest shrank, with the statistical comparison producing a probability value below 0.001. In practical terms, smaller windows tended to portray the sampled tissue as more uniform, even though the whole muscle contained a wider range of local patterns. Mean echointensity behaved in the opposite direction: it increased significantly as the region became progressively smaller. A compact window was therefore more likely to yield a brighter average signal than a larger selection. This does not necessarily mean that the muscle itself changed between measurements. Instead, it indicates that reducing the sampling area can change which tissue components dominate the calculation. If a small window happens to fall over a brighter portion of muscle, its EI may be elevated; if it misses darker or more heterogeneous areas, the resulting value may no longer describe the muscle globally.
The remaining 19 features produced more varied results, but the broad trend was consistent: measurements generally became more variable as the analyzed area decreased. This is an expected consequence of sampling statistics. A large region contains many pixels and averages over local fluctuations, reducing the influence of any single structure. A small region contains fewer observations, so its numerical features are more sensitive to the exact position of the square, the orientation of fibers, image noise and the accidental inclusion of connective tissue. Texture features derived from gray-level co-occurrence matrices and gray-level run-length matrices are especially vulnerable to this problem because they depend on the arrangement of pixels, not merely their average brightness. A shift of only a few pixels can alter the number of short or long runs, the degree of gray-level uniformity, or the frequency of neighboring intensity combinations. The smaller the window, the less stable those estimates may become.
Among the fixed-size approaches, the 300-by-300-pixel region showed the highest agreement with the whole-muscle analysis. It minimized measurement differences while also reducing the risk of including non-muscular structures that can occur when the entire visible muscle area is traced. This result does not establish that 300 by 300 pixels is a universal standard for every ultrasound machine, muscle or patient. Pixel dimensions depend on imaging settings, depth, resolution and the physical size represented by each pixel. A square of a given pixel width may therefore cover different real-world areas in different systems. The study instead points to a methodological principle: researchers need a region large enough to capture meaningful biological variation, but controlled enough to avoid contamination from adjacent tissues. In future work, that balance may be expressed in millimeters or as a percentage of muscle area rather than by pixels alone.
The implications extend beyond one calf muscle or one patient population. Quantitative muscle ultrasound is attractive in rehabilitation because it can be performed at the bedside, repeated frequently and used without ionizing radiation. Unlike computed tomography, it does not expose patients to X-rays, and unlike magnetic resonance imaging, it generally requires less equipment and can be used in settings where MRI is impractical. These advantages could make ultrasound useful for monitoring whether therapy is preserving muscle mass, reducing pathological changes or improving function. But clinical translation depends on measurements being reproducible across operators, devices and institutions. If one laboratory analyzes a whole muscle, another uses a 50-by-50-pixel window and a third selects a manually chosen patch, apparent differences may reflect image-processing decisions rather than genuine differences in muscle biology. The new findings suggest that ROI definition should be reported as carefully as the ultrasound probe, patient position and imaging protocol.
The analysis also highlights why automated and artificial-intelligence-based ultrasound tools must be trained and validated with care. Machine-learning systems can identify patterns in echotexture, but their outputs are only as meaningful as the image regions supplied to them. A model trained on small windows may learn local brightness artifacts instead of clinically relevant muscle characteristics. Conversely, a model trained on whole-muscle images may perform poorly when presented with narrowly cropped data. Standardized segmentation, repeated measurements and explicit reporting of ROI dimensions could help separate true biological signals from technical variability. The researchers’ use of agreement analysis is important in this context: correlation alone can show that two methods move in the same direction, while Bland–Altman analysis examines whether they actually produce sufficiently similar values for practical use. For a clinical biomarker, that distinction is crucial.
The study’s conclusions are appropriately focused rather than sensational: region-of-interest size substantially affects quantitative echotexture measurements, particularly EV and EI, and progressively smaller regions tend to produce greater variability and weaker agreement with whole-muscle analysis. The work was a secondary analysis of existing images from 22 post-stroke participants, so it does not by itself determine how ROI size influences diagnosis, prognosis or response to treatment. It also does not show that one measurement strategy predicts walking ability better than another. Those questions will require larger studies linking standardized echotexture features with functional outcomes and testing protocols across ultrasound systems. Still, the message is immediate and highly shareable because it exposes a hidden source of disagreement in a rapidly expanding field: in medical imaging, the answer can change depending on where—and how much—you look. By identifying the 300-by-300-pixel approach as the strongest fixed-size match to whole-muscle measurements in this dataset, the researchers provide a practical starting point for more consistent ultrasound research after stroke.
Subject of Research: The effect of ultrasound region-of-interest size on quantitative muscle echotexture measurements in the medial gastrocnemius of people after stroke
Subject of Research: Medicine
Article Title: Does size matter? A secondary analysis of the effect of region-of-interest size on muscle echotexture features of the gastrocnemius medialis in stroke
Article References: Lapuente-Hernández, D., Asadi, B., Carcasona-Otal, A., Pujol, C., Herrero, P., & Skorupska, E. (2026). Does size matter? A secondary analysis of the effect of region-of-interest size on muscle echotexture features of the gastrocnemius medialis in stroke. BMC Medical Imaging. https://doi.org/10.1186/s12880-026-02617-4
Image Credits: AI Generated
DOI: 10.1186/s12880-026-02617-4
Keywords: quantitative ultrasound, muscle echotexture, region of interest, stroke rehabilitation, gastrocnemius medialis, echointensity, echovariation, medical imaging
Cite this news
APA
MLA
Chicago
SCIENMAG. (August 28, 2026). Study tests how measurement region size affects calf muscle imaging after stroke. https://scienmag.com/study-tests-how-measurement-region-size-affects-calf-muscle-imaging-after-stroke/
SCIENMAG. “Study tests how measurement region size affects calf muscle imaging after stroke.” Scienmag, 28 August 2026, https://scienmag.com/study-tests-how-measurement-region-size-affects-calf-muscle-imaging-after-stroke/. Accessed 28 August 2026.
SCIENMAG. “Study tests how measurement region size affects calf muscle imaging after stroke.” Scienmag. August 28, 2026. https://scienmag.com/study-tests-how-measurement-region-size-affects-calf-muscle-imaging-after-stroke/
Copy citation
Download RIS
Tags: assessing medial gastrocnemius muscle post-strokecalf muscle analysiscalf muscle damage assessment post-strokeechovariation and mean echointensity in muscle analysisechovariation and mean echointensity variabilityeffects of region size on ultrasound accuracyeffects of sampling area size on muscle imaging accuracyimaging techniques for calf muscle evaluation post-strokeimpact of sampling area on muscle assessmentinfluence of region of interest selection in ultrasoundmeasurement of muscle tissue damagemeasurement variability in muscle imaging studiesmuscle structure tracking using ultrasoundmuscle tissue analysis in rehabilitationMuscle ultrasound imaging after strokenon-invasive muscle damage tracking methodsoptimizing ultrasound imaging parametersquantitative ultrasound for stroke recoveryquantitative ultrasound in stroke rehabilitationregion of interest size in ultrasoundstandardized ultrasound protocols for stroke patientsultrasound imaging techniques for muscle evaluationUltrasound measurement region size impact on calf muscle imaging after strokeultrasound-based muscle structure monitoring
