Professor Lasse Løvstakken has been awarded the NCS Research Award 2025, an accolade that recognizes a sustained record of high‑impact research at the intersection of ultrasound engineering, cardiovascular medicine, and applied machine learning. Announced in early December 2025, the award celebrates Løvstakken’s body of work — spanning novel blood flow imaging techniques, algorithmic advances for automated echocardiographic analysis, and a consistent track record of clinical translation — and highlights a larger Norwegian research ecosystem that is pushing ultrasound imaging from laboratory prototypes into everyday clinical tools. This recognition cements the role of image‑driven engineering in modern cardiology and sets a useful moment to assess both the technical accomplishments and the practical challenges that lie ahead for ultrasound‑based AI in cardiac care.
Background and overview
The NCS Research Award is presented to researchers who have contributed outstanding science to cardiology. The 2025 prize honors a researcher whose work bridges rigorous engineering with clinical utility — an ongoing theme in Løvstakken’s career. He trained as an engineer, earning an MSc in Cybernetics and a PhD in Medical Technology, and has spent the past two decades building methods and teams that translate ultrasound physics into actionable cardiac diagnostics.
Løvstakken leads the ultrasound research group at NTNU’s Department of Circulation and Medical Imaging (ISB), a team that today numbers roughly sixty scientific staff. He also played a central role in the Centre for Innovative Ultrasound Solutions (CIUS), an SFI (Centre for Research‑based Innovation) that explicitly set out to move ultrasound innovation from bench to bedside and into industry partnerships. Over his career he has authored over a hundred peer‑reviewed papers and supervised nearly thirty PhD candidates, with sixteen of those under his primary supervision. Those metrics — publications, supervision, and leadership of a major research group — provide the academic and organizational foundation that underpins the award.
What distinguishes Løvstakken’s work is not only publication volume but a consistent focus on two complementary goals: first, improving the
information content of echocardiography (for example, more precise blood‑flow vector fields and tissue motion tracking); and second, delivering
automated, robust image analysis tools powered by modern machine learning that reduce operator variability and speed clinical workflows.
Technical contributions: methods that change what echocardiography can measure
Løvstakken’s laboratory has produced multiple technical advances that extend what ultrasound systems can measure and how those measurements are used clinically. Several contributions stand out for their combination of novelty and practical impact.
Blood Speckle Imaging and Blood Speckle Tracking
One of the most visible innovations associated with Løvstakken’s group is
Blood Speckle Imaging (BSI) — a high‑frame‑rate ultrasound approach that tracks the natural speckle pattern of blood to estimate two‑dimensional velocity vector fields inside cardiac chambers. Unlike conventional Doppler, which measures velocity along the ultrasound beam and is thus angle‑dependent, speckle tracking yields a local velocity vector across the image plane. That enables visualization and quantification of vortical flow, shear, and intra‑ventricular flow patterns that were previously accessible only through complex MRI techniques or invasive measurements.
BSI and blood speckle tracking have matured from proof‑of‑concept studies into clinical case reports, pediatric applications, and validation studies that estimate intraventricular pressure differences. Notably, the method has been incorporated into commercial ultrasound platforms, which is a strong indicator of industrial uptake and practical feasibility. The pipeline from algorithm to integrated tool reflects a research strategy that emphasizes real device compatibility and clinical relevance.
Caveat: while BSI is available on specific commercial platforms and has produced promising clinical case series, claims that it is “used daily worldwide” are context‑sensitive. Adoption varies by specialty (pediatrics, congenital heart disease, specialized echo labs) and depends on probe selection, software licensing, and local training; broad, general‑purpose adoption across every echo lab has not yet occurred. The technology’s move from niche applications into routine, guideline‑driven practice will require larger multicenter outcomes data and standardized workflows.
Quantitative flow and pressure metrics
Related to BSI, Løvstakken’s group has advanced methods to estimate
intraventricular pressure differences non‑invasively from echocardiography. By combining speckle‑based flow fields with physics‑aware post‑processing, the team has produced techniques that approximate pressure gradients inside the ventricle without catheterization. Such estimates have clear clinical value: they potentially enable assessment of diastolic suction, filling dynamics, and subtle hemodynamic dysfunctions that are otherwise difficult to quantify.
These physiological measures require careful validation. The group has produced studies that compare non‑invasive estimates to invasive pressure measurements and to established imaging modalities. These validation efforts are essential because derived pressure is highly sensitive to spatial and temporal accuracy of the velocity fields; small numerical errors can amplify through the mathematical operations that transform velocity into pressure gradients.
Robust tissue tracking and deformation analysis (EchoTracker, LU‑Net, CAMUS)
On the tissue side, several machine‑learning innovations have come out of the NTNU ultrasound group that address longstanding problems in echocardiography: drift in point tracking, robustness of segmentation, and repeatability of clinical indices.
- EchoTracker is a learning‑based myocardial point‑tracking method that uses a coarse‑to‑fine model to maintain accurate point trajectories across long sequences and noisy frames. The model improves long‑range tracking and reduces drift, importantly translating into better estimates of global longitudinal strain (GLS) — a clinically important prognostic measure.
- LU‑Net and related segmentation work have improved automated delineation of left ventricular borders and chamber structures. These solutions are designed to be robust across image quality variations and have been benchmarked on large open datasets.
- NTNU researchers also contributed to the CAMUS dataset: one of the larger open‑access annotated repositories for 2‑ and 4‑chamber echocardiography. Public datasets like CAMUS enable reproducible benchmarking, accelerate algorithmic improvements, and reduce proprietary lock‑in for academic development.
The combination of open data (CAMUS), algorithmic advances (LU‑Net, EchoTracker), and device‑level adaptations (BSI) demonstrates a full‑stack approach: developing foundational algorithms, validating them on realistic data, and targeting clinical integration.
Real‑time and automated measurements
A recurring practical theme in the group’s research is
real‑time automation: neural networks and engineered pipelines that compute left ventricular ejection fraction, MAPSE (mitral annular plane systolic excursion), and wall thickness measurements in real time or near real time. Automation addresses a key clinical problem: manual measurements are time‑consuming and subject to inter‑observer variability. By lowering the measurement burden and improving reproducibility, these tools aim to speed workflow and reduce diagnostic error — critical benefits in both high‑volume clinical labs and resource‑limited settings.
Leadership, mentorship, and ecosystem building
Beyond specific algorithms, Løvstakken’s impact is magnified by institutional leadership. As head of the ISB ultrasound group, he oversees a sizeable team of researchers and students, steers collaborative projects with hospitals and industry, and contributes to large research consortia like CIUS. CIUS itself is a deliberate, long‑term program that pooled resources across academia and industry to accelerate ultrasound innovation; its final reports show hundreds of publications, dozens of patents and licensed inventions, and multiple spin‑offs or product integrations.
Academic leadership matters for translation. Training PhD students and creating reproducible data resources helps ensure that progress is maintainable after initial grants end. Løvstakken’s supervision record (more than two dozen PhDs, with many as primary supervisor) indicates a sustained commitment to capacity building — an underappreciated but critical aspect of scientific impact.
Industry partnerships and clinical translation
A defining feature of this body of work is its emphasis on translation through industry collaboration. Blood Speckle Imaging and related features are available on GE HealthCare Vivid platforms, which demonstrates that the research has met practical constraints: computational latency, probe compatibility, regulatory pathways for software features, and clinician usability.
Industry collaboration accelerates access to larger datasets, specialized transducers, and clinical sites for testing. However, industrial ties must be handled transparently. In published papers arising from these projects, disclosure statements indicate that hardware or research software keys were provided by manufacturers and that certain investigators have consultancy ties to vendors. These disclosures are standard and necessary; they allow the community to evaluate results while acknowledging shared interests.
Why the NCS Research Award matters
The award recognizes not just technical novelty but also evidence of clinical relevance and translational maturity. In a field where novel methods often remain confined to conference proceedings, receiving a national cardiology society’s research prize signals that cardiac clinicians see real potential — and sometimes real usage — in the tools developed by the research program. For the cardiovascular imaging community, this award amplifies the message that ultrasound is no longer purely structural imaging: it is increasingly a quantitative, physics‑driven modality augmented by AI.
For the broader research ecosystem, the award also highlights Norway’s model of tightly coupled research–industry–clinical partnerships, supported by programs like CIUS. This institutional context matters because building deployable medical imaging tools requires long‑term coordination across unusual technical, regulatory, and clinical domains.
Critical analysis: notable strengths
- Full‑stack engineering to clinic pipeline. The group’s work spans algorithm design, dataset creation, device integration, and clinical validation. That end‑to‑end approach reduces the friction that often prevents academic methods from reaching clinicians.
- Balance of novelty and practicality. Techniques like BSI are not just mathematical curiosities; they address concrete clinical measurement gaps (e.g., angle‑independent flow visualization and intraventricular pressure estimates).
- Open science and benchmarking. Contributions to public datasets and reproducible code bases reduce duplication of effort and enable independent validation — a key quality marker in machine‑learning research.
- Mentorship and team scale. Sustained supervision and a sizeable research group mean that progress is scalable and that knowledge is transmitted across generations of researchers.
- Industry deployment. Integration into commercial ultrasound systems demonstrates that computational requirements, ergonomics, and regulatory considerations were handled early in the development process.
Risks, limitations, and challenges
Despite clear strengths, several non‑trivial risks and limitations must be acknowledged. These are common to the field of medical imaging AI, but the specifics matter here because they affect both clinical safety and long‑term adoption.
- Generalization and data bias. Echocardiographic image quality, patient habitus, scanner settings, and population physiology vary across centers. Machine‑learning models trained primarily on data from specialized centers or particular vendor hardware can underperform when deployed in different environments. Robust external validation and prospective multicenter trials are required.
- Reproducibility under clinical variability. Flow and pressure estimations rely on precise spatiotemporal fidelity. Movement artifacts, arrhythmias, and acoustic windows with limited views can degrade accuracy substantially. Methods must be stress‑tested across arrhythmia conditions, pediatric and adult subpopulations, and varied probe geometries.
- Regulatory and medico‑legal complexity. Software features that influence diagnosis or clinical decisions are regulated as medical devices in many jurisdictions. The path to regulatory clearance requires documented clinical benefit, risk analyses, and post‑market surveillance plans. Integration with vendor ecosystems can simplify deployment, but it also creates dependencies on vendor upgrade cycles and support.
- Vendor lock‑in and accessibility. While inclusion of BSI in commercial platforms is a success for translation, dependence on specific vendors for hardware and software licenses may limit accessibility in resource‑constrained settings. Open‑source algorithms and low‑cost deployment strategies will be necessary to democratize access.
- Conflict of interest and trust. Industry collaborations accelerate development but can raise questions about independence. Transparent disclosures are necessary but insufficient: independent replication by groups without close vendor ties strengthens confidence.
- Explainability and clinician trust. Clinicians need interpretable outputs and clear failure modes. Black‑box models that output single indices without uncertainty estimates or visual validation will struggle to gain clinician acceptance. Tools that present vector fields, uncertainty maps, and clear quality metrics will be more useful.
- Clinical outcomes evidence. Improved measurements and faster workflows are important, but the field must demonstrate that these changes translate into better patient outcomes, earlier diagnoses, or cost savings. Prospective studies that link imaging enhancements to clinical endpoints remain the gold standard.
Practical recommendations and what to watch next
For technologists, clinicians, and administrators evaluating these innovations, the near term should focus on rigorous, multi‑center validation and practical deployment steps.
- Prioritize external validation across vendor hardware and diverse clinical settings to reduce the risk of dataset bias and overfitting.
- Require safety‑oriented reporting: publish failure modes, uncertainty quantification, and standardized quality metrics alongside performance numbers.
- Favor open datasets and reproducible code; open benchmarks accelerate progress and reduce duplication.
- Insist on clinical utility studies that measure impact on workflow, diagnostic confidence, and, where feasible, patient outcomes.
- Keep transparency about industry relationships and promote independent replication by unaffiliated groups.
Key items to watch in the coming years include broader availability of vector flow imaging on mainstream echo platforms, prospective trials that evaluate speckle‑derived pressure metrics for diastolic dysfunction and heart failure management, and regulatory approvals that define safe use cases for AI‑enabled echo features.
Contextual perspective: translating innovation into everyday care
The arc of innovation visible in Løvstakken’s work — from algorithmic prototypes to vendor‑supported features — illustrates the maturation of ultrasound research over the last decade. Echocardiography has historically been qualitative and heavily operator‑dependent. The combined application of high‑frame‑rate imaging, physics‑informed processing, and modern machine learning creates a new class of
quantitative echocardiography: reproducible, validated measurements that can be standard parts of the cardiac exam.
This shift is not purely technical. It requires rethinking training for sonographers and cardiologists, updating reporting standards, and evolving regulatory frameworks. The reward is potentially vast: more precise hemodynamic assessment at the bedside, better triage of acute conditions, and affordable, radiation‑free modalities that can be repeated frequently without patient harm.
Conclusion
The NCS Research Award 2025 presented to Professor Lasse Løvstakken recognizes a coherent program of research that has advanced both the theoretical and practical state of echocardiography. His group’s work — spanning Blood Speckle Imaging, machine‑learning driven tissue tracking, robust segmentation, and device integration — demonstrates how focused engineering can expand the diagnostic possibilities of an established modality.
At the same time, the next phase of impact hinges on rigorous external validation, transparent disclosures of vendor relationships, and clinical studies that demonstrate not only better measurements but better outcomes. The award is therefore both a recognition of past achievement and a call to the community: scale these technologies responsibly, measure their real‑world benefits, and ensure they are accessible beyond specialized centers. If those conditions are met, the innovations celebrated by this award have the potential to materially improve cardiac diagnostics and patient care worldwide.
Source: Oslo Universitetssykehus
NCS Research Award 2025 to Lasse Løvstakken