Fractional flow reserve, or FFR, has been the reference standard for judging the functional significance of a coronary narrowing since the concept was formalized in the early 1990s and hardened through landmark randomized trials. The index is elegantly physical: it expresses the maximum blood flow an artery can deliver through a stenosis as a fraction of the flow the same artery would deliver if the vessel were perfectly open. Because flow during maximal hyperemia is proportional to perfusion pressure, FFR is measured invasively as the mean pressure distal to the narrowing divided by the mean pressure in the aorta, with hyperemia induced by intravenous or intracoronary adenosine. A value of 0.80 or below — meaning the stenosis consumes at least a fifth of the pressure that should be driving the heart muscle’s blood supply — defines a hemodynamically significant lesion. Randomized trials such as FAME and FAME 2 established that guiding stent decisions by this pressure ratio, rather than by the eye, reduces adverse cardiac events and spares patients unnecessary implants, and international guidelines have endorsed the measurement for exactly the kind of intermediate lesions that plague daily practice.

Yet the technique remains underused relative to its evidence base. Crossing the lesion with a dedicated pressure guidewire adds procedural time and cost; adenosine frequently causes flushing, chest burning, breathlessness and transient conduction abnormalities that patients find deeply unpleasant; and the measurement demands technical care, because wire handling, catheter damping and inadequate hyperemia can all distort the result. Meanwhile, decades of comparison studies have shown that visual estimation from angiography is a poor surrogate for physiology: roughly half of angiographically intermediate lesions turn out to be non-flow-limiting when actually measured, and experienced operators frequently disagree with one another about severity. This is the gray zone in which the “oculostenotic reflex” — the reflexive impulse to stent anything that looks tight — flourishes, contributing both to the over-treatment of lesions that would never have caused ischemia and to the occasional under-treatment of those that would. The gap between what the artery looks like and what it does is precisely the gap that computational physiology was invented to close.

Vessel fractional flow reserve belongs to the fast-growing family of angiography-derived, computed FFR indices — cousins of the CT-derived FFR-CT that pioneered the field. The software ingests standard coronary angiograms acquired from at least two projection angles and reconstructs a three-dimensional model of the arterial lumen, quantifying the vessel’s caliber, length and lesion geometry along its full course. Blood flow is then estimated from angiographic frame counts and vessel dimensions, and the pressure drop across the reconstructed geometry is solved numerically with computational fluid dynamics — in most implementations, variants of the Navier–Stokes equations that govern pressure and flow in a tube — accounting for viscous friction losses along diffuse disease and for separation losses at focal narrowings. The output is a hyperemic pressure ratio calculated without a wire, without adenosine and within minutes — and, because the computation is distributed along the entire artery, a pressure profile showing exactly where along the vessel flow is lost, which is the origin of the term “vessel” FFR.

Optical coherence tomography attacks the same stenosis from the opposite direction: it ignores hemodynamics entirely and interrogates the structure of the vessel wall. The technique works like ultrasound with light, using a fiber-optic imaging catheter that emits near-infrared radiation at a wavelength of roughly 1.3 micrometers; because red blood cells scatter light ferociously, blood is transiently flushed from the artery with contrast during the acquisition. Second-generation frequency-domain systems pull back at high speed, generating cross-sectional images with an axial resolution of 10 to 20 micrometers — roughly ten times finer than intravascular ultrasound. At that scale, operators can measure the minimal lumen area, map the arc of lipid-rich plaque, quantify the thickness of its fibrous cap, and identify macrophage infiltration, cholesterol crystals, neovascular microchannels, thrombus and the depth of calcification. Where FFR answers whether a stenosis matters, OCT answers what the stenosis is made of — information that has become indispensable both in assessing vulnerable plaque and in optimizing stent implantation.

VISION-FFR brought the two technologies together in a single-center, prospective, observational design. The researchers enrolled patients with chronic coronary syndromes — the umbrella term for the stable and chronic presentations of coronary artery disease, as opposed to acute heart attacks — whose angiograms showed intermediate stenoses of 40 to 80 percent. Every participant underwent the full workup: wire-based FFR with adenosine, computed vFFR derived from the angiogram, and intracoronary OCT of the target vessel, so that 120 lesions in 106 patients could each be characterized simultaneously in physiological and anatomical terms. Because all three indices were assessed in the same lesions, the design eliminates the lesion-selection drift that plagues comparisons across separate cohorts. A cutoff of 0.80 defined hemodynamic significance for both indices. The cohort split cleanly along that line: sixty-two patients fell into the vFFR-positive group, with a median vFFR of 0.72 (interquartile range 0.70–0.77), while forty-four patients were vFFR-negative, with a median of 0.88 (interquartile range 0.85–0.92) — a separation that was statistically significant and consistent with a real physiological divide between the two populations.

The study’s distinctive ambition lies in what that divide might be made of. The stated objective was to evaluate the association between vFFR and OCT-derived parameters — to test whether computed physiology tracks with plaque anatomy measured at micron resolution, and whether lesions that computationally strangle flow also display the OCT signatures cardiologists associate with obstructive and high-risk morphology, such as small minimal lumen areas, large lipid arcs and thin fibrous caps. Previous efforts to predict wire-based FFR from anatomy alone have delivered mixed results: thresholds based on minimal lumen area measured by intravascular imaging improve on visual estimation but still misclassify a substantial share of lesions, because the functional impact of a narrowing depends not only on its tightest point but also on lesion length, on the abruptness of its entry and exit, on diffuse disease elsewhere in the artery and on the mass of heart muscle it supplies. By capturing the computed pressure field and the microscopic plaque landscape in the same lesions, VISION-FFR assembles precisely the paired dataset needed to probe why some anatomies translate into ischemia while others do not.

The practical stakes are considerable. If computed vFFR proves reliable across larger populations, the pressure wire and the adenosine infusion could be reserved for the minority of cases in which the computation is untrustworthy — poor image quality, heavy calcification, complex bifurcations or left main disease — shortening procedures, cutting costs and sparing patients the drug’s side effects. Because vFFR is derived from angiograms that are acquired anyway, it can even be computed retrospectively on stored images, potentially flagging lesions that deserve formal physiological testing. And in the hybrid workflow the Warsaw team’s design anticipates, OCT and vFFR are natural companions: the imaging catheter is already in the artery, and its micron-scale anatomy can contextualize a pressure number that the wire alone cannot explain — distinguishing, for instance, a focal lipid-rich culprit from diffuse disease producing the same value. The caveats are real, though. The study was single-center and observational, with a moderate sample size and no long-term outcomes; computed indices inherit every artifact of the angiograms they are built from; and OCT, for all its resolution, does not quantify diffuse flow limitation along the artery’s length the way a pullback pressure trace does.

The report lands amid a broader migration of coronary physiology from the catheter to the computer. CT-derived FFR, validated in large multicenter studies more than a decade ago, demonstrated that hemodynamic significance could be simulated noninvasively and even used to defer invasive angiography altogether in stable patients; angiography-derived platforms followed, with validation studies against wire-based FFR showing high agreement, and machine-learning implementations have since compressed computation times from hours to seconds. The economic logic is equally attractive: a computation that runs on images already acquired costs a fraction of a disposable pressure wire, and an index that needs no adenosine removes both the drug’s discomfort and the minutes spent waiting for hyperemia. What VISION-FFR adds is the anatomical cross-examination: rather than asking only whether the computed number agrees with the wire, the Polish team asked what the plaque looks like when the algorithm declares a vessel positive, anchoring the virtual physiology in physical structures that operators can see, measure and — eventually — target.

Larger, multicenter and ideally outcome-driven studies will be needed before computed vessel physiology and OCT-derived morphology can jointly decide who receives a stent and who goes home on medication alone. But the direction of travel is unmistakable. Cardiology has spent a generation learning that the angiogram’s grayscale silhouette is an unreliable narrator of ischemia, and the tools that correct it are becoming faster, cheaper and image-only. If that evidence matures, the catheterization laboratory of the near future could look radically leaner: an angiogram, a computation and, where anatomy demands explanation, a light-based pullback. If subsequent work confirms the Warsaw findings, the intermediate stenosis may finally lose its status as the catheterization laboratory’s most persistent dilemma — adjudicated instead by a digital twin of the coronary circulation, cross-examined by light.

Cardiovascular disease is often managed through snapshots. A patient may undergo a scan, receive a diagnosis and then return months later for another examination, while important biological changes occur between visits. Intravascular imaging offers a closer look by placing miniature imaging devices inside blood vessels. Technologies such as intravascular ultrasound and optical coherence tomography can reveal the structure of vessel walls, plaque deposits and implanted stents with far greater detail than many external imaging methods. The article presents these tools as the foundation for a more responsive form of care, in which information from inside the vessel can guide decisions with greater precision.

Photonics is central to this vision because light can carry detailed information about tissue composition and microscopic structure. Optical coherence tomography, for example, uses reflected near-infrared light to produce high-resolution cross-sectional images, allowing clinicians to examine thin fibrous caps, small dissections and the surfaces of stents. Other optical approaches, including near-infrared spectroscopy and photoacoustic imaging, can add chemical or molecular information. Photoacoustic systems work by delivering short pulses of light and detecting the acoustic waves created when tissue absorbs that energy. In principle, combining these signals could help distinguish stable plaque from lesions more likely to rupture.

The challenge is that vascular disease rarely follows the boundaries of a single organ or one anatomical location. Atherosclerosis can affect coronary, carotid, renal, peripheral and cerebral vessels at the same time, while systemic inflammation, diabetes, high blood pressure and abnormal lipid metabolism influence the entire circulation. This “panvascular” perspective is important because treating one narrowed segment may not address the processes driving disease elsewhere. The article argues that intelligent photonics could help link local images with broader physiological data, creating a more complete picture of vascular health rather than focusing only on the most visible obstruction.

Artificial intelligence could make this expanding stream of information clinically useful. Intravascular scans contain enormous numbers of pixels and complex patterns that may be difficult to interpret consistently, even for experienced specialists. Machine-learning systems can be trained to identify vessel boundaries, measure plaque volume, classify tissue characteristics and evaluate the position or expansion of stents. When combined with patient histories, laboratory measurements, medication records and blood-flow simulations, these algorithms could support risk assessment and help physicians compare changes across repeated examinations. The goal is not simply automated image reading, but the integration of diverse data into clinically meaningful predictions.

At the heart of the proposed framework is the digital twin: a computational representation of an individual patient that is updated as new measurements become available. In engineering, digital twins are used to monitor machines and predict failures. In medicine, a vascular digital twin could combine three-dimensional anatomy, blood-flow dynamics, tissue properties and biological risk factors. Computational fluid dynamics could estimate how blood moves through narrowed or branching vessels, while imaging data could refine the model’s geometry. As new scans or physiological measurements arrive, the virtual representation could be adjusted, allowing clinicians to explore how a disease might progress or how a proposed intervention could alter circulation.

Such a system could eventually support adaptive vascular care, in which treatment changes according to the patient’s evolving condition rather than following a fixed schedule. A digital model might help assess whether a plaque is becoming more dangerous, whether a stent is causing abnormal flow, or whether medication is reducing the biological activity associated with disease. It could also provide a framework for testing possible interventions virtually before they are performed. However, these possibilities depend on reliable data, validated algorithms and careful clinical oversight. A prediction generated by software would need to be tested against real outcomes before it could be trusted in routine care.

The article also highlights major barriers between an attractive concept and a usable medical platform. Imaging systems must become faster, safer and easier to operate, while the data they generate must be standardized across hospitals and manufacturers. Artificial intelligence models need large, diverse and well-annotated datasets so that they do not perform well only on the patients used during development. Privacy and cybersecurity are essential because a digital twin would contain highly sensitive medical information. Most importantly, prospective clinical studies must determine whether these technologies actually improve diagnosis, treatment decisions and patient outcomes, rather than merely producing more detailed images.

The emerging message is that vascular medicine may be moving from episodic diagnosis toward continuous, data-driven monitoring. Intelligent photonics could reveal what is happening inside vessels at microscopic and molecular scales, while artificial intelligence could organize those observations and digital twins could place them into a personalized physiological model. The paper does not present this future as a finished clinical reality; instead, it offers a roadmap for connecting imaging, computation and patient care across the vascular system. If the technical and ethical challenges can be solved, the result could be a new generation of medicine that detects danger earlier, adapts treatment more precisely and views circulation as one interconnected living network.

Subject of Research: Intelligent photonics, intravascular imaging, artificial intelligence and digital twins for adaptive care in panvascular disease

Article Title: From intravascular imaging to adaptive vascular care: intelligent photonics and digital twins in panvascular disease

Article References: You, L., Yao, J., Qiu, Y. et al. From intravascular imaging to adaptive vascular care: intelligent photonics and digital twins in panvascular disease. Light Sci Appl 15, 335 (2026). https://doi.org/10.1038/s41377-026-02410-6

Image Credits: AI Generated

DOI: 10.1038/s41377-026-02410-6

Keywords: Intravascular imaging, photonics, digital twins, artificial intelligence, vascular disease, atherosclerosis, adaptive medicine, cardiovascular technology

Optical coherence tomography has long been a powerful imaging tool during catheter-based cardiac interventions such as angioplasty and stent placement. Despite its unparalleled capability to render high-resolution images revealing the detailed structure of blood vessels, conventional OCT imaging lacks the biochemical specificity required to discern the composition of vessel walls. This limitation impedes cardiologists’ ability to fully assess the vulnerability of plaques to rupture, a critical factor in predicting heart attack risk.

The research team, led by Hyeong Soo Nam from the Korea Advanced Institute of Science and Technology (KAIST), has devised a new approach that harnesses wavelength-dependent characteristics embedded in OCT signals. By integrating these spectral insights with advanced artificial intelligence, their system can non-invasively detect and map the distribution of lipid deposits inside coronary arteries. This ability to identify lipid content provides a previously inaccessible level of detail crucial for evaluating patient risk.

Published in Biomedical Optics Express, the study details a sophisticated methodology for extracting subtle spectral information from standard OCT images. Unlike traditional modifications requiring specialized hardware, this AI-driven solution works seamlessly with the OCT systems already deployed in clinical settings. It reflects a major innovation in computational imaging, leveraging deep learning for automated, quantitative tissue characterization without additional equipment costs or procedural changes.

This AI-powered advancement is poised to enhance clinical decision-making during coronary interventions. By offering real-time, objective data on lipid presence, the tool can aid physicians in assessing risks more accurately, tailoring procedural strategies, and monitoring treatment responses. The ultimate benefit lies in enabling individualized patient care plans that reduce the likelihood of adverse cardiac events and improve long-term health outcomes.

A core technical achievement of the research is the sophisticated extraction and analysis of spectral data from OCT signals, which IIllustrates tissue-specific light-tissue interactions. Lipids, fibrous tissue, and calcifications each exhibit distinct optical absorption and scattering properties across different wavelengths of light. The AI model effectively learns to detect these unique patterns, enabling an automated and robust identification of lipid-rich plaque areas throughout the vessel wall.

This approach uniquely combines weakly supervised deep learning with spectroscopic OCT. Importantly, it reduces the annotation burden that often hampers AI model training. Instead of requiring detailed pixel-level annotations of lipid regions—an arduous and subjective manual task—the system learns from simpler frame-level labels indicating the presence or absence of lipids. This strategy enhances practicality and scalability, facilitating real-world clinical adoption.

To validate their model’s accuracy and clinical relevance, the team applied the method to intravascular imaging data from a rabbit model of atherosclerosis. They rigorously compared the AI-derived lipid detection outcomes against conventional histopathology using lipid-specific staining techniques. The results demonstrated high accuracy in classifying lipid presence and strong spatial correspondence between AI-highlighted regions and histologically confirmed lipid deposits.

The research heralds a new era in the application of AI to intravascular imaging. Beyond OCT, the framework offers potential for extension to other optical or vascular imaging modalities where subtle spectral variations remain underutilized. This adaptability suggests a broad future impact, encouraging the development of AI-integrated diagnostic tools for a variety of cardiovascular diseases and other pathologies.

Looking forward, the team is focused on optimizing the system for speed and robustness, key factors for implementation in the fast-paced clinical environment. Further validation with human coronary artery data will be crucial to confirm translatability and determine best practices for integration into existing clinical workflows. Ensuring that the technology complements physician workflows without disruption will be critical to its adoption and success.

This innovative AI method represents a substantial leap forward in cardiovascular diagnostics, pairing the sophisticated physics of spectroscopic OCT with state-of-the-art computational techniques. The ability to non-invasively, accurately, and rapidly detect lipid-rich plaques offers a powerful new tool in combating the global burden of heart disease. With further development, it has the potential to save countless lives through earlier intervention and personalized treatment.

The study not only exemplifies the promise of AI-enhanced medical imaging but also underscores the importance of multidisciplinary collaboration—in this case, merging expertise in optical physics, clinical imaging, pathology, and machine learning. Such convergences are driving the future of precision medicine, enabling physicians to unlock new dimensions of insight from existing diagnostic technologies.

For clinicians and researchers alike, this work marks a pivotal step towards safer, more effective management of coronary artery disease. As healthcare increasingly embraces AI-driven innovations, tools like this herald a transformative shift towards predictive, preventive, and personalized care—factors essential to addressing one of the leading causes of global mortality.

Subject of Research: Artificial intelligence-based detection of lipid-rich plaques within coronary arteries using spectroscopic optical coherence tomography.

Article Title: Automated lipid detection in spectroscopic optical coherence tomography using a weakly supervised deep learning network.

News Publication Date: Information not provided.

Web References:
– Biomedical Optics Express journal: https://www.osapublishing.org/boe/home.cfm
– DOI link: https://opg.optica.org/boe/abstract.cfm?doi=10.1364/BOE.585222
– KAIST: https://www.kaist.ac.kr/en/

References:
J. H. Hwang, W. Lee, J. H. Kim, R. H. Kim, D.O. Kang, J. W. Kim, H. Yoo, H. S. Nam, “Automated lipid detection in spectroscopic optical coherence tomography using a weakly supervised deep learning network,” Biomed. Opt. Express, 17, 1279-1292 (2026). DOI: 10.1364/BOE.585222

Image Credits: Hyeong Soo Nam, Korea Advanced Institute of Science and Technology

Keywords

Artificial intelligence, Cardiac arrest, Optical coherence tomography, Medical imaging