CARA Lab -- Cardiology Lab with Abbott, Radboudumc, and AmsterdamUMC
CARA Lab researches automated segmentation and characterization of intravascular structures and lesions, optimizing OCT-based algorithms to assess physiologic characteristics, high-risk plaque identification, and predicting stent-related complications, with the goal of increasing the usability, reliability, and applicability of intravascular OCT in interventional cardiology through the use of trustworthy AI. The lab is a collaboration between Abbott, Radboud University Medical Center, and Amsterdam University Medical Center.
CARA Lab aims to address two challenges that are common in many real-world AI applications: (1) developing accurate AI models that can analyze high-resolution data streams in real-time and (2) deriving accurate and safe models from these data streams that can make long-term predictions with a limited number of training samples. Our focus is on using optical coherence tomography (OCT) pullbacks, which are high-resolution data streams that are critical for decision-making during cardiac interventions. Our goal is to identify patients at risk for future cardiac events so that preventative treatment and enhanced monitoring can be applied to those individuals. We aim to substantially enhance the usability, reliability, and applicability of intravascular OCT in interventional cardiology through the use of trustworthy AI.
To achieve these goals, CARA Lab is focusing on automated segmentation and characterization of intravascular structures and lesions, optimizing OCT-based algorithms to assess physiologic characteristics, high-risk plaque identification, and predicting stent-related complications. Indicators here are publications about the accuracy and repeatability of the developed algorithms. We consider algorithms accurate when they fall within the interobserver variability of human analysts in core laboratory assessment.
The major assumption is that AI-aided interpretation of OCT images will result in increased use of OCT in revascularization decision-making, which will be tested through the evaluation of patient outcomes and surveys. Another assumption is that the OCT interpretation by a core laboratory used to develop and validate the algorithms is accurate. Initial research showed that there is a good overall correlation between OCT findings and histopathological findings, but there are some limitations that can be addressed through comparison with other imaging modalities such as computed tomography (CT). It is also assumed that treating vulnerable plaques could have a clinical benefit for patients, though this has not yet been formally demonstrated in a clinical trial.
CARA Lab will test the impact of AI-driven OCT on (1) decision-making and (2) trust within the catheterization laboratory for both physicians and patients.
CARA Lab is part of the ROBUST program on Trustworthy AI-based Systems for Sustainable Growth which is financed under the NWO LTP funding scheme. While contributing to SDGs 3 (Good Health) and 9 (Innovation), we aim to support interventional cardiology by optimizing procedural decision-making based on OCT imaging supported by AI algorithms for contemporary indications, as well as for new indications. As such, we enhance the potential of OCT as an all-in-one tool in interventional cardiology, aiming at improving procedural outcomes and ultimately patient outcomes.
Optical coherence tomography (OCT) is a relatively new imaging modality within interventional cardiology and cardiovascular research. Previous research has demonstrated its potential in several key aspects of the management of patients with atherosclerotic coronary artery disease, leading to beneficial patient outcomes. Currently, its use in daily clinical practice is limited due to a lack of training for the assessment and analysis of OCT-acquired pullbacks. We aim at easing the use of OCT by means of artificial intelligence (AI), thereby increasing the availability of AI-based OCT algorithms for clinical decision-making, and eventually improving patients’ health, thereby contributing to Sustainable Development Goal 3.
Besides aiming at developing real-time AI support to make OCT easier to use during interventions, we aim at developing new AI algorithms to improve and extend the current use of OCT imaging, to determine patients at risk of new cardiac events, and to extract hemodynamic information. This has the ability to further enhance the potential of OCT to become an all-in-one tool in interventional cardiology, contributing to Sustainable Development Goal 9.
Target 3.4: By 2030, reduce by one third premature mortality from non-communicable diseases through prevention and treatment and promote mental health and well-being
Target 9.5: Enhance scientific research, and upgrade the technological capabilities of industrial sectors in all countries, in particular developing countries, including, by 2030, encouraging innovation and substantially increasing the number of research and development workers per 1 million people and public and private research and development spending
OCT obtains a large amount of high-resolution cross-sectional frames. Interpretation requires extensive training and time and it is questionable whether a physician is able to correctly interpret all of the acquired information on-site, which is required for ad-hoc decision-making in acute coronary syndromes. Accurate interpretation includes distinguishing different plaque types and identifying acute plaque complications including plaque rupture and thrombus formation. Real-time automated annotation has the potential to speed up and improve the accuracy of OCT interpretation. Moreover, although it is known that several plaque characteristics are associated with adverse outcomes, the presence of these features does not account for all future events nor does the absence of these features indicate freedom from adverse events. Probably yet unknown characteristics provide prognostic and predictive value that can be discovered using AI-based approaches.
Furthermore, current computed flow reserves are based on luminal measurements and bifurcation fractional laws. However, plaque characteristics (e.g., irregular surfaces) potentially influence the physiological impact of stenosis. Integration of these characteristics in computed flow reserves may improve the accuracy of such indices. Even so, although its incidence has decreased, stent thrombosis is still one of the most dramatic clinical presentations in patients who underwent percutaneous revascularization. Other reasons for stent failure are in-stent restenosis or persistent flow limitation leading to ischemia and chest pain. Understanding the underlying cause can help guide iterative treatment.
Lastly, the impact of AI-driven OCT on the guidance of percutaneous revascularization can be questioned. Firstly, it is currently unknown how patients and clinicians would react to AI-driven decision-making in the catheterization laboratory and what potential shortcomings need to be overcome for the safe and accepted use of AI in that clinical scenario. Secondly, it is unknown how AI-driven OCT guidance impacts clinical outcomes. Although previous studies have highlighted the potential advantages of OCT guidance, a prospective study including AI-driven OCT guidance is currently lacking.
Abbott, the Radboud University Medical Center, and the Amsterdam University Medical Center, will all be responsible for obtaining sufficient data for the development and validation of the intended algorithms. Development, testing, and validation will mostly be performed by members of the radiology and cardiology departments of the Radboud University Medical Center and Amsterdam University Medical Center. Abbott will have the leading role in implementing the final product within a readily available OCT system. To enable this, Abbott will also be involved in the development phase by means of half-yearly meetings and all researchers will spend time at Abbott R&D in Norwood, Massachusetts, e.g. via internships.