An Introduction to AI in Cardiovascular Imaging
1. Artificial Intelligence in Cardiovascular Imaging: JACC State of the Art Review
For an introduction to AI in cardiovascular imaging, Damini Dey et al. provide a comprehensive review that covers key terms, opportunities, and challenges in the publication “Artificial Intelligence in Cardiovascular Imaging: JACC State-of-the-Art Review.” The article begins with an introduction to key terms and concepts related to AI, helping the reader understand the three steps to use AI in imaging: 1.) Data and Features, 2.) Computational Techniques, and 3.) Imaging applications.
- Disease phenotyping & cluster analysis, which provides a process of creating related groups from hidden patterns within data, may provide opportunities to better characterize disease.
- Diagnostic support provided by automated segmentation of anatomical structures and automated measurements.
- Image interpretation, including the utilization of imaging databases to develop deep learning-based systems.
This publication is a strong, comprehensive reference that can help facilitate the reader’s understanding of concepts in medical imaging AI.
AI-based Prediction of Plaque Progression
2. Machine Learning Framework to Identify Individuals at Risk of Rapid Progression of Coronary Atherosclerosis: From the PARADIGM Registry
A recent technical publication written by Donghee Han et al. explains how machine learning algorithms could be used to identify patients at risk of rapid progression of coronary atherosclerosis. In the publication titled “Machine Learning Framework to Identify Individuals at Risk of Rapid Progression of Coronary Atherosclerosis: From the PARADIGM Registry,” Han et al. employed a machine learning-based prediction model that would automatically perform qualitative and quantitative plaque characterization from 1,073 patients from the Progression of Atherosclerotic Plaque Determined by Computed Tomography Angiography Imaging (PARADIGM) Registry. The machine learning model integrated clinical, laboratory, and computed tomography-based qualitative and quantitative plaque features. Before the development of this model, no method existed for identifying individuals at risk of rapid plaque progression at a single point in time.
The study findings demonstrated that CCTA features, such as those from quantitative plaque analysis, were the most important factors in determining plaque progression. The machine learning-based model also showed a higher predictive performance for rapid plaque progression as compared to the combination of using clinical and lab, and qualitative CT-based variables. This shows that the prediction model could facilitate the ability to identify patients at risk of future plaque progression without requiring invasive testing. This publication supports the utilization of machine learning models for automatic risk calculation to better inform care decisions.
Deep Learning-based CT-FFR
3. Influence of Coronary Stenosis Location on Diagnostic Performance of Machine Learning-based Fractional Flow Reserve from CT Angiography
Deep learning-based computed tomography-derived fractional flow reserve (CT-FFR) is a non-invasive technique in which FFR values are calculated from CCTA images, helping patients to avoid unnecessary invasive testing. Numerous recent studies have shown that deep learning-based CT-FFR techniques can reduce the need for invasive coronary angiography and improve outcomes in patients with stable coronary artery disease. In a new study published in the Journal of Cardiovascular Computed Tomography titled, “Influence of Coronary Stenosis Location on Diagnostic Performance of Machine Learning-based FFR from CT Angiography,” Matthias Renker et al. investigated the effect of stenosis location on a machine learning-based CT-FFR method.
Cardiovascular Disease Risk Prediction
4. Deep Convolutional Neural Networks to Predict Cardiovascular Risk from Computed Tomography
Several recent studies have introduced how AI-assisted opportunistic screening patients at risk of cardiovascular disease can help clinicians harness additional information from routine chest CT scans. In a new publication available in Nature Communications, Zeleznik et al. developed a deep learning model that could automatically calculate coronary artery calcium scores from cardiac CT scans and chest CT scans in under two seconds without human input. The study titled, “Deep Convolutional Neural Networks to Predict Cardiovascular Risk from Computed Tomography” included a total of 20,084 individuals. Asymptomatic patients from the National Lung Screening Trial (NLST) and the Framingham Heart Study (FHS) were included in the study, as well as patients with stable and acute chest pain from the Prospective Multicenter Imaging Study for Evaluation of Chest Pain (PROMISE) and the Rule Out Myocardial Infarction Using Computed Assisted Tomography (ROMICAT II) trials. In four consecutive steps, a coronary calcium risk score was calculated fully automatically through heart localization, heart segmentation, calcium segmentation, and coronary artery calcium score.
AI-assisted Opportunistic Screening for Cardiovascular Disease
5. Automated Coronary Calcium Scoring Using Deep Learning with Multicenter External Validation
Building on the research from Zelzenik et al., recent research published in June demonstrates the added value that deep learning-based algorithms can bring in facilitating opportunistic screening from un-enhanced chest CT exams. In the NPJ Digital Medicine article titled, “Automated Coronary Calcium Scoring Using Deep Learning with Multicenter External Validation,” David Eng et al. describe how they developed a fully automatic, end-to-end deep learning model for coronary artery calcium scoring using gated coronary CT and non-gated routine un-enhanced chest CT exams. The deep learning model could automatically calculate coronary artery calcium scores from dedicated gated coronary CT exams and routine non-gated chest CT scans performed for other reasons to allow for opportunistic screening.
Clinicians’ Role in Cardiovascular Imaging AI to Clinical Practice
6. Machine Learning and the Future of Cardiovascular Care: JACC State-of-the-Art Review
While AI and other advanced informatics can help solve real clinical challenges in cardiovascular imaging, it will require clinicians’ active involvement to bring valuable algorithms into clinical use. Giorgio Quer et al. provide a framework to help clinicians evaluate new developments in AI in the publication, “Machine Learning and the Future of Cardiovascular Care: JACC State-of-the-Art Review.” The authors begin by providing an overview of six machine learning concepts that clinicians need to be familiar with to effectively read research publications on machine learning methods. These six concepts include:
- Machine learning algorithms learn rules and patterns from data that is fed to them, rather than having to be explicitly programmed like traditional rules-based algorithms.
- Supervised learning algorithms can learn patterns from labeled data.
- Machine learning algorithms can learn patterns without labeled examples through unsupervised learning and reinforcement learning.
- Overfitting is a common problem in machine learning and occurs when algorithms perform well on training data but fail on test data.
- Accuracy, interpretability, and explain-ability of a machine learning algorithm.
- Machine learning algorithms can be retrained to include more data or different data types.
 Zeleznik, Roman & Foldyna, Borek & Eslami, Parastou & Weiss, Jakob & Alexander, Ivanov & Taron, Jana & Parmar, Chintan & Alvi, Raza & Banerji, Dahlia & Uno, Mio & Kikuchi, Yasuka & Karady, Julia & Zhang, Lili & Scholtz, Jan-Erik & Mayrhofer, Thomas & Lyass, Asya & Mahoney, Taylor & Massaro, Joseph & Vasan, Ramachandran & Aerts, Hugo. (2021). Deep convolutional neural networks to predict cardiovascular risk from computed tomography. Nature Communications. 12. 10.1038/s41467-021-20966-2.
 Quer G, Arnaout R, Henne M, Arnaout R. Machine Learning and the Future of Cardiovascular Care: JACC State-of-the-Art Review. J Am Coll Cardiol. 2021 Jan 26;77(3):300-313. doi: 10.1016/j.jacc.2020.11.030. PMID: 33478654; PMCID: PMC7839163.