The ultrasound probe glides across the patient's neck, its grainy black and white images flickering on the screen. To most observers, these moving pictures reveal little more than shadowy shapes. But hidden within those shadows lies information that could mean the difference between preventing a stroke and suffering one. The challenge has always been extracting that information quickly and accurately enough to guide treatment decisions.

For decades, doctors have relied on ultrasound imaging to examine the carotid arteries, the major blood vessels in the neck that supply blood to the brain. When fatty deposits called plaques build up in these arteries, they can break apart and trigger strokes. The question that haunts every physician examining these scans is deceptively simple: which plaques are dangerous, and which can be safely monitored?

Now, researchers in Cyprus have developed software that can answer this question by analyzing not just what plaques look like, but how they move.

The Deceptive Calm of Asymptomatic Disease

Consider the peculiar challenge facing vascular surgeons. A patient walks into the clinic feeling perfectly healthy. An ultrasound reveals significant narrowing of the carotid artery, perhaps 70% blocked by plaque. Should the surgeon operate?

Surgery carries real risks. Carotid endarterectomy, the procedure to clean out these plaques, comes with a 2% to 3% chance of causing the very stroke it aims to prevent. For patients already experiencing symptoms like mini strokes, this risk is acceptable because their danger of a major stroke without surgery is much higher. But for asymptomatic patients, those feeling fine despite their narrowed arteries, the calculus becomes agonizing.

Studies have shown that in asymptomatic patients with severe narrowing, surgery reduces the annual stroke risk from about 2% down to 1%. That sounds good until you remember the surgery itself carries a 2% to 3% immediate risk. The math suggests that most asymptomatic patients might be better off without surgery, at least initially. But which ones? How do you identify the subset of asymptomatic patients whose plaques are ticking time bombs?

This is where texture, composition, and motion enter the story.

What a Plaque Reveals

Atherosclerotic plaques are not uniform blobs. They have complex internal structures: fibrous caps, lipid cores rich in cholesterol, areas of calcification as hard as bone, regions of dead tissue, sometimes even ulcers that look like tiny craters. These different components show up differently on ultrasound images, creating textures that trained eyes can interpret.

A plaque with a large lipid core covered by a thin fibrous cap is dangerous. It resembles a water balloon with a weak spot, likely to rupture. A heavily calcified plaque, while still problematic, tends to be more stable. Between these extremes lie countless variations, each with its own risk profile.

For years, researchers have worked to quantify these visual differences using computer analysis. They developed methods to measure image texture, identifying patterns associated with high-risk versus low-risk plaques. Gray scale median values, texture uniformity, the presence of discrete white areas or juxtaluminal black areas, all these features have proven useful in predicting outcomes.

But static images only tell part of the story. Your arteries are not frozen in place. With every heartbeat, they expand and contract. The plaque moves with them, or sometimes, it does not move quite as it should.

The Motion That Matters

During each cardiac cycle, as your heart pumps blood through the carotid arteries, the arterial walls expand slightly during systole when pressure is highest and relax during diastole. A stable plaque moves in harmony with this rhythm, all its parts moving together in what researchers call concordant motion.

A vulnerable plaque behaves differently. Its various components lipid, fibrous tissue, calcification may move asynchronously, sliding against each other, creating internal stresses. This discordant motion suggests instability, like feeling slight movement in a supposedly solid structure that warns of impending collapse.

Previous studies had shown that this motion could be captured and quantified, but the process was laborious. Researchers had to painstakingly analyze video recordings frame by frame, tracking how different parts of the plaque moved throughout the cardiac cycle. It was research work, not something that could be done routinely in a busy clinic.

The team in Cyprus, building on years of prior research, set out to change that.

Building AtheroRisk

The software they developed, called AtheroRisk, brings together multiple strands of prior research into a single integrated system. It handles the entire workflow from loading an ultrasound video to producing a stroke risk assessment.

The process starts with preprocessing. Raw ultrasound videos vary in quality, resolution, and intensity calibration depending on the machine settings and technique. AtheroRisk standardizes these videos, normalizing them to a consistent resolution of 20 pixels per millimeter. It adjusts image brightness using reference points the user selects in areas of blood and bright arterial wall to ensure comparable measurements across different scans.

The software can also reduce speckle noise, the grainy appearance inherent to ultrasound imaging that can obscure subtle features. Users choose between filtering methods that have been previously validated for this specific application.

Next comes identification of the cardiac cycle. Not all moments in a heartbeat are equal. Early systole, middle systole, peak systole, and diastole each offer different views of plaque behavior. AtheroRisk generates what is called a motion mode image, essentially stacking vertical slices from each video frame horizontally to create a visualization of movement over time. This allows users to identify and mark different cardiac phases, defining exactly which portions of the video will be analyzed.

Seeing Structure, Tracking Movement

With the cardiac phases identified, AtheroRisk moves to segmentation, outlining the plaque boundaries. Users can draw these manually with pixel level precision, or employ the software's deep learning model for automated segmentation. This model, adapted from recent advances in medical image analysis, has been trained to recognize plaque boundaries across different plaque types, from uniformly dark lipid rich plaques to heavily calcified bright lesions.

Once the plaque is outlined, AtheroRisk extracts dozens of features. Geometric measurements include total plaque area, perimeter, major and minor axes. Texture analysis employs multiple mathematical approaches, each capturing different aspects of the image patterns: first order statistics describe basic intensity distributions, gray level dependence matrices quantify spatial relationships between pixels, run length features characterize continuous stretches of similar brightness.

Some of these features have proven particularly valuable. Gray scale median, for instance, correlates with plaque composition: echolucent (dark) plaques with low values tend to be lipid rich and dangerous, while echogenic (bright) plaques with high values contain more fibrous tissue and calcium. The presence of juxtaluminal black areas, dark regions adjacent to the vessel lumen without a visible echogenic cap, independently predicts stroke risk.

Beyond static features, AtheroRisk analyzes motion using optical flow methods. It tracks how each part of the plaque moves between consecutive video frames throughout the cardiac cycle. The software calculates the dominant orientation of motion and, crucially, measures angular spread.

In a concordant plaque, motion vectors point mostly in the same direction, producing a narrow angular spread. In a discordant plaque, different regions move in different directions, creating a wide angular spread that can exceed 120 degrees or even 340 degrees in extremely unstable cases.

From Features to Risk

All these measurements feed into risk stratification models developed from large databases of patients whose outcomes were tracked for years. The Asymptomatic Carotid Stenosis and Risk of Stroke study followed over 1,000 patients with significant asymptomatic narrowing for up to eight years, documenting which features predicted strokes.

AtheroRisk combines clinical information like degree of stenosis and patient symptoms with image features and motion characteristics to produce two key outputs: a five year stroke free survival rate and an average annual stroke risk rate.

These are not just academic numbers. They guide treatment decisions. A patient with low risk scores might be managed medically with medications and lifestyle changes. A patient with high scores despite being asymptomatic might be referred for surgery or stenting to prevent a future stroke.

Testing the System

To verify that AtheroRisk works as intended, the research team analyzed 54 ultrasound videos: 27 from asymptomatic patients and 27 from symptomatic patients who had already experienced strokes or transient ischemic attacks.

The results showed clear separation between the groups. Maximum angular spread of motion differed significantly, with a p value of 0.006, indicating this finding would occur by chance less than once in 150 trials. The median motion magnitude also differed significantly, with p equal to 0.032.

Texture features similarly distinguished the groups. First order statistics measuring asymmetry in brightness distribution, spatial gray level features quantifying image uniformity, and gray level dependence features all showed different patterns between asymptomatic and symptomatic plaques.

These findings aligned with previous research, confirming that the integrated software produces reliable results. More importantly, AtheroRisk can generate these results in real time during a clinical examination, something no previous tool could accomplish.

The View Inside

Looking at actual examples helps illustrate what AtheroRisk reveals. In one case, an asymptomatic patient aged 68 with 95% carotid narrowing showed concordant plaque motion. The motion analysis displayed a maximum angular spread of just 59 degrees, with most plaque components moving together toward the lower right during middle systole. The polar diagram showed a tight cluster of motion vectors, the scatterplot revealed consistent direction across different motion magnitudes.

Contrast this with a discordant case where the maximum angular spread reached 341 degrees. The polar diagram showed motion vectors pointing in nearly opposite directions, the scatterplot scattered chaotically. Different parts of the plaque, particularly the darker echolucent regions, moved independently, creating internal shear forces that stressed the plaque structure.

The software can also generate pseudo colored visualizations mapping plaque composition. Dark blue regions represent lipid cores, green and yellow show fibrous tissue, orange and red indicate calcification. These color maps make the internal structure immediately apparent, helping physicians understand why a particular plaque behaves as it does.

Advanced analysis using amplitude modulation and frequency modulation techniques reveals even finer details of texture at multiple scales, though these analyses require more computational time and are used selectively for research purposes or complex cases.

Beyond the Research Lab

The team designed AtheroRisk with clinical workflow in mind. All patient information is automatically anonymized, assigned random identifiers to protect privacy. Data gets stored in a local database using SQLite, allowing physicians to track patients over time, compare serial scans, and document disease progression or regression in response to treatment.

The software handles multiple video formats, though the developers recommend lossless compression to preserve image quality. It can even convert DICOM sequences, the standard medical imaging format, into video files for analysis.

Everything runs on a standard computer without requiring specialized hardware. The interface guides users through each step, from loading a video to generating a final report. Experienced ultrasonographers can complete an analysis in minutes once they become familiar with the workflow.

Early feedback from physicians who tested the software at conferences has been positive, though formal validation with larger numbers of users is ongoing. The team plans a six month trial where vascular specialists will use AtheroRisk in actual clinical practice, analyzing plaques and generating risk scores as part of routine care. Their experience will guide refinements before the software seeks regulatory approval for widespread clinical use.

Limitations and Future Directions

No tool is perfect, and the researchers acknowledge current limitations. Cases with multiple plaques require analyzing each separately. Plaque ulcers, crater like defects that significantly increase stroke risk, cannot always be identified from ultrasound alone, sometimes requiring additional imaging.

The deep learning segmentation model, while generally accurate, sometimes struggles with unusual plaque geometries or heavy calcification that creates acoustic shadows. In these cases, manual segmentation by experienced users produces better results.

Future versions will address these challenges. The team is working on improved algorithms for detecting ulcers based on motion patterns, reasoning that the complex flow dynamics around ulcers might create characteristic movement signatures. They are also developing methods to automatically detect and separately analyze multiple plaques in a single video.

Integration with other imaging modalities represents another frontier. Combining ultrasound motion analysis with CT angiography or MRI could provide even more detailed characterization of high risk plaques.

The Bigger Picture

AtheroRisk exemplifies a broader trend in medical imaging: moving beyond simple visualization to quantitative analysis. Radiology is becoming increasingly computational, with algorithms extracting information that human eyes cannot readily perceive.

This raises questions about the future role of human expertise. Will software replace the judgment of experienced physicians? The developers say no. AtheroRisk is designed as a decision support tool, providing objective measurements to inform clinical judgment, not replace it. The physician still interprets results in context, considers patient preferences, weighs risks and benefits.

But the tool does democratize expertise to some extent. A physician using AtheroRisk in a small clinic can access analytical capabilities previously available only at major research centers. This could improve care in underserved areas where specialist expertise is scarce.

Broader Implications for Prevention

The stakes extend beyond individual patients. Stroke remains a leading cause of death and disability worldwide. Even non-fatal strokes often leave survivors with permanent impairments affecting speech, movement, and cognition. The economic costs, including medical care and lost productivity, run into billions annually.

Better stroke prevention could dramatically reduce this burden. But prevention requires identifying high risk individuals before they have strokes. For carotid disease, this means distinguishing which asymptomatic plaques need aggressive intervention and which can be managed conservatively.

Current guidelines rely heavily on degree of stenosis, the percentage of arterial narrowing. But stenosis alone is an imperfect predictor. Some patients with 80% stenosis never have strokes, while others with 60% stenosis do. Adding plaque characterization and motion analysis to the assessment could significantly improve risk stratification.

Studies suggest that features like juxtaluminal black areas, discrete white areas, and echolucency correlate strongly with stroke risk independent of stenosis severity. Motion analysis adds another dimension. If these features can be routinely assessed in clinical practice, treatment decisions become more evidence based and personalized.

The Patient Perspective

From a patient's viewpoint, this technology offers both promise and complexity. Imagine being told you have significant carotid narrowing but no symptoms. Do you choose surgery with its immediate risks, or watchful waiting with the uncertainty of future stroke?

Now imagine your doctor can show you detailed analysis of your plaque: its composition, its stability, how it moves with your heartbeat. The doctor explains that your plaque shows concordant motion, minimal lipid core, heavy calcification. Your five year stroke risk is 3%. Surgery would reduce this to perhaps 1.5%, but carries a 2.5% immediate risk. The choice becomes clearer: watchful waiting with aggressive medical management makes sense.

Alternatively, your analysis reveals discordant motion, large lipid core, juxtaluminal black areas. Your five year stroke risk is 18%. Now surgery, despite its risks, looks more attractive. The discussion shifts from "should we operate" to "when should we schedule surgery."

This kind of quantitative, personalized risk assessment empowers both physicians and patients to make informed decisions aligned with individual circumstances and values.

Validation and Regulatory Hurdles

Before AtheroRisk can enter routine clinical use, it must clear significant hurdles. Regulatory agencies like the European Medicines Agency and US Food and Drug Administration require extensive validation demonstrating that the software is safe and effective.

This means more than showing it can distinguish asymptomatic from symptomatic plaques in retrospective data. It means prospective studies where physicians use the software to guide actual treatment decisions, with long term follow up documenting outcomes. Do patients assessed as low risk by AtheroRisk truly have low stroke rates? Do high risk assessments justify intervention?

Such studies take years and significant resources. They must enroll hundreds or thousands of patients, follow them for years, meticulously document every stroke, TIA, and procedure. Only with this evidence can regulators and clinicians be confident the software delivers on its promise.

The research team is preparing for this journey, seeking partnerships with hospitals and vascular centers willing to participate in validation trials. They aim for CE marking in Europe within three years, a milestone that would allow clinical use across the European Union.

A Tool for Research

Beyond clinical applications, AtheroRisk serves research purposes. Understanding plaque biology requires studying many examples, correlating imaging features with composition verified through pathology when plaques are surgically removed, tracking how plaques evolve over months and years.

Manual analysis of hundreds of ultrasound videos is prohibitively time consuming. Automated analysis through AtheroRisk makes large scale studies feasible. Researchers could, for instance, investigate whether certain medications preferentially stabilize plaques by reducing lipid content or improving fibrous cap strength, tracking changes in texture features and motion patterns over time.

The software's database capabilities facilitate this research. Every analysis generates structured data: measurements, features, classifications. Aggregate this across many patients and patterns emerge that would be invisible in individual cases.

Technical Innovation Under the Hood

While users see a streamlined interface, significant computational sophistication operates behind the scenes. The optical flow algorithms tracking motion must handle frame to frame variations in image quality, compensate for patient movement, distinguish plaque motion from arterial wall motion.

The deep learning segmentation model was trained on thousands of manually annotated images representing diverse plaque types and imaging conditions. Training such models requires careful curation of data, augmentation techniques to expand the training set, validation strategies to avoid overfitting.

The texture analysis employs multiple mathematical frameworks, each with theoretical foundations in image processing and pattern recognition. Gray level dependence matrices, for instance, build on information theory principles. Run length features derive from texture perception models. Each captures different aspects of image structure, and combining them provides richer characterization than any single approach.

The software architecture balances several competing demands. It must process video data efficiently, performing complex calculations in near real time. It must handle data securely, protecting patient privacy through anonymization and encryption. It must provide an intuitive interface despite underlying complexity.

Meeting these requirements demanded careful software engineering, iterative development with user feedback, extensive testing across different platforms and use cases.

The Collaborative Foundation

AtheroRisk did not emerge from nowhere. It synthesizes decades of research by multiple groups. Some developed the image standardization methods. Others validated texture features against clinical outcomes. Still others pioneered motion analysis techniques.

The Cyprus team's contribution was integration: bringing these elements together in a cohesive, usable system. They added pieces where gaps existed, refined methods to work robustly in practice, designed the overall workflow for clinical utility.

This exemplifies how medical technology often advances. Rarely does a single brilliant insight revolutionize practice. More often, progress comes from careful engineering that translates research findings into practical tools, combining proven techniques in novel ways.

The collaborative nature extends to ongoing work. The team partners with vascular clinics collecting ultrasound data, with universities developing machine learning methods, with hospitals conducting validation studies. Each contributor adds expertise the others lack.

Looking Ahead

The future roadmap for AtheroRisk includes several directions. Enhanced automation to handle more complex cases with minimal user input. Integration with electronic health records to streamline clinical workflows. Mobile versions running on tablets for point of care use.

The team envisions eventually connecting AtheroRisk with national registries tracking stroke outcomes. This would enable continuous validation and refinement as the software accumulates real world experience across diverse populations and practice settings.

There are also aspirations to extend beyond carotid disease. Similar plaque analysis principles could apply to peripheral artery disease affecting leg arteries, or coronary artery disease affecting heart arteries. The software architecture is designed for adaptability.

Artificial intelligence offers another frontier. Current deep learning models handle segmentation, but future versions might predict outcomes directly from raw images, learning risk patterns too subtle or complex for handcrafted features to capture. Early experiments show promise, though substantial work remains.

The Human Element

For all its sophistication, AtheroRisk ultimately serves a profoundly human purpose: helping people avoid strokes. Behind every analyzed video is a person, often unaware they carry disease, hoping to stay healthy long enough to watch grandchildren grow, to finish careers, to enjoy retirement.

The software does not make these people's anxieties disappear. Learning you have significant carotid disease is frightening even when risk scores are reassuring. But it can transform vague fear into concrete information that guides action.

Perhaps most importantly, it shifts the paradigm from "wait and see if you have a stroke" to active prevention based on individual risk. This aligns with broader movements in medicine toward personalized, predictive approaches that intervene before disease causes harm.

The physicians using AtheroRisk appreciate this shift. Rather than relying on crude rules like "operate if stenosis exceeds 70%," they can tailor recommendations to each patient's unique situation. This feels more like true medical practice, applying science to serve individual human needs.

A Window on Tomorrow

AtheroRisk represents what medical imaging is becoming. The future of radiology is not just better pictures, but smarter analysis extracting actionable information from those pictures. Algorithms will increasingly work alongside human expertise, handling quantitative measurements while physicians provide judgment, context, and human connection.

This future carries challenges. Ensuring algorithms work fairly across diverse populations. Maintaining skills as automation handles routine tasks. Addressing liability when software contributes to decisions. Preserving the human element in increasingly technological medicine.

But it also carries immense promise. Better diagnosis, earlier intervention, more personalized care, democratized expertise. Tools like AtheroRisk move us toward this future, one ultrasound video at a time.

For the patient in the examination room, ultrasound probe against their neck, the promise is simpler but no less profound: the chance to see danger before it strikes, to act before it is too late, to maintain the health and independence we all treasure. That is a promise worth pursuing.

Publication Details

Year of Publication: 2025 (online available)
Journal: SN Computer Science
Publisher: Springer Nature
DOI Link: https://doi.org/10.1007/s42979-025-03666-2

About This Article

This article is based on original peer-reviewed research published in SN Computer Science. All findings, technical specifications, and conclusions presented here are derived from the original scholarly work. This article provides a simplified overview for general readership. For complete methodological details, comprehensive technical specifications, validation procedures, statistical analysis, and full academic content, readers are strongly encouraged to access the original research article by clicking the DOI link above. All intellectual property rights belong to the original authors and publisher.

Medical Disclaimer

This article is provided for informational and educational purposes only and does not constitute medical advice, diagnosis, or treatment. The research described reflects early-stage laboratory findings, and the experimental nanoneedle gene-editing technology discussed is not approved for clinical use. Readers should not make healthcare decisions or alter treatment plans based on this content and should consult qualified healthcare professionals for personalized medical guidance. No doctor–patient relationship is created by this article, and professional medical advice should always be sought for any health concerns or emergencies.