Part 2 of 5 in Circle's Coronary Plaque series. Also read: Part 1 — How Advanced Plaque Analysis Changes the Clinical Calculus When clinical cardiology adopts a new capability, IT inherits the infrastructure. And right now, coronary plaque analysis is moving from research tool to clinical standard fast enough that many IT and PACS teams are still catching up. The demand is real. The 2021 ACC/AHA Chest Pain Guidelines made CCTA a Class I recommendation for stable chest pain evaluation. More recent trial data — including 10-year outcomes from SCOT-HEART and the ongoing SCOT-HEART 2 trial — is driving cardiology programs to go further, adding quantitative plaque characterization alongside standard stenosis reporting. That means new software, new data flows, new integrations — and new complexity landing in your environment. How that complexity lands depends almost entirely on the path the department chooses. There are essentially two: a unified platform that performs plaque analysis natively, inside your existing environment — or a send-away service that moves CCTA data out of your network to a vendor cloud, runs the analysis there, and returns a result. Those two paths lead to very different IT outcomes.
Part 1 of 5 in Circle's Coronary Plaque series. Consider a familiar scenario. A 52-year-old presents with atypical chest discomfort. The stress test is borderline. The referring physician is uncertain — is this true coronary disease, or something else? Standard imaging has been done. The anatomy looks, on the surface, mostly normal. And yet something about this patient doesn't sit right. This is the case where clinicians have historically had to make consequential decisions with incomplete information. Escalate to invasive angiography and risk an unnecessary procedure. De-escalate and risk missing a vulnerable plaque that hasn't yet caused significant stenosis — but will. Coronary artery disease doesn't announce itself neatly. A substantial share of acute coronary events occur in patients with non-obstructive coronary disease — lesions that would not have triggered revascularization on a standard angiogram. The stenosis grade has always been an imperfect surrogate for risk. What matters is the plaque itself: its composition, its burden, and its vulnerability. Advanced CCTA plaque analysis changes that calculus.
The Effort, Risks, and Why It’s Worth It Quantifying the Effort of a Single Platform Learning a new (or deeper) way of working Even for existing cvi42 users, full standardization across cardiac imaging means: Adapting to new modules or modalities. Rethinking how you structure reading lists when more work is visible in one place. Helping develop shared templates and protocols. It’s completely normal to see a temporary dip in productivity when something new is introduced. Our brains are wired for loss aversion: we tend to focus more on what we might lose than on what we could gain. Recognizing that bias helps put the slowdown in perspective: it’s not a setback, it’s just part of the process that leads to long‑term progress.
Local Processing for Complete Data Security When cvi 42 processes imaging data, everything takes place within the customer’s secure environment. All image data and derived results are managed locally, whether on a hospital workstation or through a customer-managed server installation. No data is ever transmitted outside the institution. This architecture ensures compliance with strict hospital IT policies and data protection frameworks. For clinical users, this means AI-powered results without any compromise to data privacy or network security. The Circle AI Engine: Trained, Validated, and Frozen “Each of the AI models powering cvi42 is architected and developed within Circle’s controlled research and development environment. Circle’s data science and clinical AI research teams use diverse and representational datasets to train and validate each algorithm. The process typically involves supervised learning, where the AI learns to recognize patterns and structures such as the left ventricle, myocardium, or aortic root by comparing its results to expert-annotated data. Once performance meets clinical and regulatory standards, the AI model is locked, “frozen” and encrypted during its integration within cvi42. This means the model’s behavior is fixed, it does not continue to learn or change once deployed at a customer site. The model you use in cvi 42 is the validated version approved for clinical use, ensuring consistent and reproducible results across all installations. No Learning from Customer Data It is important to clarify: the AI in cvi 42 does not learn from any data processed at the customer site. The algorithm applies its pre-trained parameters to each image set locally. It does not store patient data, send information externally, or modify its internal model based on what it sees or whether a user edits its outputs. Each analysis is isolated, ensuring the AI’s decisions remain consistent and the patient’s information stays protected within the facility’s network. How the AI Analyzes Medical Images At a technical level, cvi42’s AI is a deep learning-based image analysis engine trained to recognize and segment cardiac anatomy on MR and CT images. Primarily using convolutional neural networks, it performs pixel- or voxel-level classification to delineate key structures, including the endocardial and epicardial borders. These segmentations enable the measurements of clinically relevant metrics such as chamber volumes, ejection fraction, and myocardial mass. This process mimics how expert readers would interpret the same dataset, but it happens in seconds and with objective consistency across cases. Designed for Trust, Built for Performance AI in cvi 42 is designed to automate routine analysis while keeping clinicians fully in control. Users can review, adjust, and approve AI-generated contours as needed, ensuring that results always meet their clinical standards. Combined with local data processing, frozen AI models, and Circle’s rigorous training pipeline, this approach delivers accuracy and reliability without ever compromising patient privacy.
