JACC: CARDIOVASCULAR INTERVENTIONS VOL. 12, NO.

14, 2019

ª 2019 PUBLISHED BY ELSEVIER ON BEHALF OF THE

AMERICAN COLLEGE OF CARDIOLOGY FOUNDATION

STATE-OF-THE-ART REVIEW

Impact of Artificial Intelligence on

Interventional Cardiology
From Decision-Making Aid to Advanced Interventional
Procedure Assistance

Partha Sardar, MD,a J. Dawn Abbott, MD,a Amartya Kundu, MD,b Herbert D. Aronow, MD,a Juan F. Granada, MD,c
Jay Giri, MD, MPHd,e

ABSTRACT

Access to big data analyzed by supercomputers using advanced mathematical algorithms (i.e., deep machine learning)
has allowed for enhancement of cognitive output (i.e., visual imaging interpretation) to previously unseen levels and
promises to fundamentally change the practice of medicine. This field, known as “artificial intelligence” (AI), is making
significant progress in areas such as automated clinical decision making, medical imaging analysis, and interventional
procedures, and has the potential to dramatically influence the practice of interventional cardiology. The unique nature of
interventional cardiology makes it an ideal target for the development of AI-based technologies designed to improve
real-time clinical decision making, streamline workflow in the catheterization laboratory, and standardize catheter-based
procedures through advanced robotics. This review provides an introduction to AI by highlighting its scope,
potential applications, and limitations in interventional cardiology. (J Am Coll Cardiol Intv 2019;12:1293–303)
© 2019 Published by Elsevier on behalf of the American College of Cardiology Foundation.

A rtificial Intelligence (AI) is a general term that

signifies the use of mathematical algorithms
which give machines the ability to reason
and perform cognitive functions such as problem
solving, object/word recognition and
making (1,2). AI encompasses a family of operations
including machine learning (ML), deep learning
(DL), natural language processing (NLP), cognitive
computing, computer vision and robotics (Figure 1)
that can be used to integrate and interpret complex
biomedical data as well as advance technological
automation (3–6). Years of research have finally
brought AI into our daily lives with popular practical
decision- applications such as self-driving cars and speech
recognition software such as Amazon’s Alexa Voice
(Amazon, Seattle, Washington) and IBM’s Watson
(IBM, Armonk, New York). However, the rate of adop-
tion of AI in medicine is slow, as compared with other

From the aCardiovascular Institute, Warren Alpert Medical School at Brown University, Providence, Rhode Island; bDivision of
Cardiovascular Medicine, University of Massachusetts Medical School, Worcester, Massachusetts; cCardiovascular Research
Foundation, Columbia University Medical Center, New York, New York; dPenn Cardiovascular Outcomes, Quality and Evaluative
Research Center, Leonard Davis Institute of Health Economics, University of Pennsylvania, Philadelphia, Pennsylvania; and the
e
Cardiovascular Medicine Division, University of Pennsylvania, Philadelphia, Pennsylvania. Dr. Granada has received institutional
grant/research support (to Skirball Center for Innovation) from Abbott Vascular, Amaranth Medical, Amber Medical, Amgen,
Baylis, BIO2 Medical, Bristol-Myers Squibb, Boston Scientific, Cagent Vascular, Caliber Therapeutics, Cephea, Columbia Medical,
Corindus Vascular, Celyad, Freudenberg Medical, Intact Vascular, JenaValve, Keystone Heart, LimFlow Medical, LoneStar Heart,
Marvel Medical, Medtronic, Meril Life Sciences, MicroVention, Motus GI, Navigate Cardiac Structures, New York University,
OrbusNeich Medical, SoundBite Medical, Spectranetics, Toray Industries, Vetex Medical, Volcano (Philips), and Zimmer Biomet.
Dr. Giri has served on an advisory board for AstraZeneca; and has received research support to the institution from Recor Medical
and Abbott Vascular. All other authors have reported that they have no relationships relevant to the contents of this paper to
disclose.

Manuscript received August 8, 2018; revised manuscript received February 26, 2019, accepted April 2, 2019.

ISSN 1936-8798/$36.00 https://doi.org/10.1016/j.jcin.2019.04.048

1294 Sardar et al.
Artificial Intelligence and Interventional Cardiology
ABBREVIATIONS industries such as financial technology, in-
AND ACRONYMS
formation technology, and the aerospace
industry.
AI = artificial intelligence
The application of AI in interventional
AR = augmented reality
cardiology (IC) can be divided into 2 main
DL = deep learning
branches, virtual and physical. The virtual
EMR = electronic medical
branch includes informatics from ML/DL,
record
NLP, and cognitive computing to control
FDA = Food and Drug
health management systems (i.e., electronic
Administration
health records and medical image analysis
IC = interventional cardiology
software), and automated clinical decision
ML = machine learning
support systems. The physical branch is best
NLP = natural language
represented by robotic interventional pro-
processing
cedures. Applications of AI within medicine
PCI = percutaneous coronary
intervention have been predicted to fundamentally
VR = virtual reality
transform the landscape of health care de-
livery. However, as with many emerging
technologies, the true promise of AI within medicine
at large, and IC specifically, may be lost if it is not
developed correctly.
CONCEPTS UNDERLYING AI METHODS
AI is an umbrella term for a set of algorithms that
imitate intelligent human behavior. ML refers to an
automated system that learns to perform a task or
make decisions automatically from an available data
source, rather than having its behavior explicitly
programmed (Table 1). Although the terms artificial
intelligence and machine learning are often used more
or less synonymously, in more precise terms, ML is a
set of techniques to enable AI. AI and ML are used
in several overlapping fields such
cognitive computing, computer vision, and robotics
(Central Illustration). Major techniques/approaches in
AI and ML include supervised, unsupervised, and
reinforcement learning (Figure 1).
ML methods can complement and extend con-
ventional statistical methods. Existing statistical
methods primarily help to explore relationships
between a limited number of variables. By contrast,
ML methods identify features from
perform prediction, and provide tools and algo-
rithms to understand patterns from large, complex,
and heterogeneous data (7,8). Additionally, ML al-
gorithms are based on fewer assumptions and can,
in some cases, provide superior and more robust
predictions (2). For example, Azzalini et al. (9) used
generalized boosted regression (an ML approach) to
identify whether contrast media type was an inde-
pendent predictor of contrast-induced acute kidney
injury after percutaneous coronary intervention
(PCI).
JACC: CARDIOVASCULAR INTERVENTIONS VOL. 12, NO. 14, 2019
JULY 22, 2019:1293–303
HIGHLIGHTS

The field of AI has initiated a paradigm
shift in health care with advanced
analytical techniques.

Potential applications of AI in IC are
image and video analysis, clinical deci-
sion support, robotic assistance with
procedure, and novel approaches to
clinical database analysis.

The current development of AI in IC is in
its early stage, but it has the potential to
transform IC practice.
DL is a subfield of ML and is characterized by al-
gorithms that are inspired by the workings of the
human brain, including a class of algorithms called
neural networks (1,7) (Table 1, Figure 1). In the case of
traditional ML, an algorithm needs to be programmed
how to make a precise prediction by providing it with
additional information. With DL, algorithms analyze
large datasets, automatically discover patterns, and
learn without human intervention. A DL neural
network consists of digitized inputs, such as an im-
age, audio or video, which proceeds through multiple
layers of connected “neurons” that progressively
identify features, and ultimately provides an output
(1,2). DL is the underlying platform for image recog-
nition applications that are likely to be used in car-
diovascular imaging modalities (e.g., angiography,
as NLP, echocardiography, computed tomography, cardiac
magnetic resonance, intravascular ultrasound, and
optical coherence tomography) (7,8). Creating DL
networks can take a significant amount of time and
resources. The transfer learning approach, which in-
volves transferring the knowledge gained by a DL
system in 1 dataset to another dataset with different
data, may mitigate this issue. (4).
the data, SUPERVISED LEARNING. Supervised learning is a
type of ML in which training of datasets are con-
ducted with specific labels or annotations (Figure 1).
This approach develops a model to predict or classify
future events, or to find variables relevant to a spec-
ified outcome. It often involves classification of an
observation into one or more categories or outcomes
(e.g., “Does this angiogram represent coronary
dissection or thrombosis?”). Supervised learning in-
volves classification and regression problems, but it
requires a lot of data and is time-consuming, because
the source data have to be labeled by humans (2,6,7).
JACC: CARDIOVASCULAR INTERVENTIONS VOL. 12, NO. 14, 2019
JULY 22, 2019:1293–303
F I G U R E 1 Types of Machine Learning
However, some successes have already been seen
with supervised learning with models developed
to successfully improve heart failure readmission
prediction and accurately grade echocardiographic
mitral regurgitation (8).
UNSUPERVISED LEARNING. Unsupervised learning
seeks to discover underlying structure or relation-
ships among variables in a dataset (Figure 1). The
training of dataset is conducted without any spe-
cific labels, and the algorithm clusters data to
reveal underlying patterns. It seeks to identify
novel disease mechanisms, genotypes, or pheno-
types from hidden patterns present in the data
(2,7). Unsupervised learning algorithms including
artificial neural networks that analyzed surface
echocardiograms have already been found to be
useful in the automated discrimination of hyper-
trophic cardiomyopathy from physiological hyper-
trophy seen in athletes (10). Shah et al. (11) used
agglomerative hierarchical clustering, a commonly
used unsupervised learning tool for analysis of
phenotypic data, and the phenomapping resulted in
a novel classification of heart failure with preserved
ejection fraction.
REINFORCEMENT LEARNING. Reinforcement learning,
based on behavioral psychology, uses an alternate
approach where a software program acts in a
Sardar et al. 1295
Artificial Intelligence and Interventional Cardiology
pre-specified environment to maximize a reward
(1,2) (Figure 1). The program thus identifies the
appropriate behavior using a “reward criteria” to
influence the decision-making process. It attempts to
accomplish a task (e.g., driving a car, inferring
medical decisions) while learning from its own suc-
cess and failures (2,6). The main aim of reinforce-
ment learning is to maximize the accuracy of
algorithms using trial and error. Many clinical
problems can be formatted to fit the format of a
reinforcement learning problem. Hence, reinforce-
ment learning algorithms may be used to aid clinical
decision making, intelligently segment medical im-
aging data, and select personalized medications
(1,2). Application of reinforcement learning to med-
icine and cardiology thus far has been limited.
Studies have demonstrated promising results with
reinforcement learning in optimization of treatment
decisions for chronic illnesses and recommending
mechanical ventilator weaning protocols that have
led to superior clinical outcomes (7).
NATURAL LANGUAGE PROCESSING. NLP, a conflu-
ence of AI and linguistics, focused upon developing a
computer’s ability to understand human language (6)
(Central Illustration, Table 1). ML and DL have some
overlap with NLP; however, NLP has a strong lin-
guistics component (not represented in images) that
requires an understanding of how we use language.
1296 Sardar et al.
Artificial Intelligence and Interventional Cardiology
T A B L E 1 Commonly Used Terms in AI
AI A branch of applied computer science wherein computer algorithms are
trained to perform tasks, and have the capability to imitate intelligent
human intelligence.
ML Subfield of AI, a machine learns to perform a task or make decisions
automatically from an available data source without being explicitly
programmed.
DL DL is a type of ML that mimics the operation of the human brain and
includes a class of algorithms called neural networks.
NLP NLP is an area of computer science and artificial intelligence related to the
organization of unstructured narrative text into a structured form that
can be interpreted by a machine and allows for automated information
extraction.
Cognitive Cognitive computing platforms integrate machine learning, reasoning,
computing natural language processing, speech and object recognition,
human–computer interaction, dialog and narrative generation.
Computer vision Computer vision is a branch of computer science concerned with objects
and feature recognition in images or multi-dimensional data, including
digital video frames.
Robotics Robotics deals with the design, construction, operation, and use of robotic
devices that can move and react to sensory input. Robotics also
concerned with creation of computer systems for their control, sensory
feedback, and information processing.
AI ¼ artificial intelligence; DL ¼ deep learning; ML ¼ machine learning; NLP ¼ natural language processing.
NLP has been used for largescale database analysis of
the electronic medical record (EMR) to detect adverse
events and post-operative complications from physi-
cian documentation, to achieve automated claims
coding, and to improve workflow (6,12). Sohn et al.
(13) demonstrated the feasibility of NLP techniques in
processing radiology reports and developed a rule-
based algorithm to identify abdominal aortic aneu-
rysm. NLP can be used to scan thousands of data
sources (EMRs, image libraries, and so on) to screen
and identify patients with critical valvular or vascular
stenosis, hypertrophic cardiomyopathy, atrial septal
defects, and so on.
COMPUTER VISION. Computer vision is a branch of
computer science dealing with objects and feature
recognition in digital images, including digital video
frames (1,5). It is a combination of image processing
(for feature extraction) plus ML (using these features
to build a model). Unsupervised learning algorithms
including different neural network can be used for
model building. An important health care–related
application of computer vision is acquisition and
interpretation of cardiac images, including computer-
aided diagnosis and image-guided procedures/
surgery. A real-time analysis of laparoscopic videos
yielded 92.8% accuracy in automated identification of
the steps of sleeve gastrectomy and also successfully
identified missing or unexpected steps (5). Although
predictive video analysis is in its infancy, evidence
exists that AI can be leveraged to process large
amounts of interventional data to identify or predict
JACC: CARDIOVASCULAR INTERVENTIONS VOL. 12, NO. 14, 2019
JULY 22, 2019:1293–303
real-time adverse events and assist intraprocedural
clinical decision making.
VIRTUAL APPLICATION OF AI IN IC
IMAGING INTERPRETATION. AI promises to have a
major impact on imaging reconstruction, analysis and
interpretation. The analysis of large amounts of im-
aging data and the use of dedicated imaging software
has allowed the advancement of the imaging display
into more “anatomic layouts” allowing the simplifi-
cation of imaging interpretation (4,8). A study by Du
et al. (14) showed the preliminary feasibility of DL
technology in angiographic recognition based on a
convolutional neural network model. For training
data, 3,990 images were used to establish the model,
and 2,711 images were used to evaluate it. The preci-
sion (percentage of the correct pixels) and recall rate
(percentage of retrieved relevant pixels) were used to
evaluate the effectiveness of the model. For detecting
lesion characteristics, including diameter stenosis,
calcification, thrombus, and dissection, the recall rate
was 88.2%, 82.6%, 85.5%, and 85.8%, respectively
(14). Another study by Ciusdel et al. (15), evaluated an
AI-based solution for fully automated cardiac phase
and end-diastolic frame detection on coronary an-
giograms. The workflow was trained with a deep
neural network on 17,800 angiograms from 3,900
patients and evaluated on 27,900 angiograms from
6,250 patients. Cardiac phase detection had an accu-
racy of 92.6%, a sensitivity of 92.4% and a specificity
of 92.9% on the evaluation set (15). A group from
Emory University has developed an ML method for
segmenting intravascular ultrasound images and
automatically calculating lumen area and plaque
burden that showed excellent agreement with an
expert analyst (16). This method can segment single
images in a fraction of a second and has the potential
for online use in the catheterization laboratory.
The anatomic and functional assessment of coro-
nary stenosis is now possible with the use of nonin-
vasive imaging using DL (HeartFlow, Redwood City,
California) (17). In the catheterization lab, computa-
tional fluid dynamics has allowed the development of
wire-free algorithms for the detection of functionally
significant coronary stenoses with use of computer
vision and 3-dimensional reconstruction (angiog-
raphy-based fractional flow reserve; Cathworks Ltd,
Kfar-Saba, Israel) (18). Recently, Siemens Healthi-
neers received Food and Drug Administration
(FDA) clearance for TrueFusion, a cardiovascular
application that integrates advanced ultrasound and
angiographic imaging for improved navigation
and guidance during structural heart disease
JACC: CARDIOVASCULAR INTERVENTIONS VOL. 12, NO. 14, 2019 Sardar et al. 1297
JULY 22, 2019:1293–303 Artificial Intelligence and Interventional Cardiology

C ENTR AL I LL U STRA T I O N Applications of AI in IC

Sardar, P. et al. J Am Coll Cardiol Intv. 2019;12(14):1293–303.

Applications of artificial intelligence (AI) includes AI-guided diagnosis, image interpretation, clinical decision support, data driven therapeutics, research
and development, population health, efficient administration, workflow and regulation, and AI-assisted interventional procedures. CT-FFR ¼ computed
tomography fractional flow reserve; EHR ¼ electronic health record; IVUS ¼ intravascular ultrasound; OCT ¼ optical coherence tomography.

interventions (19). This system integrates the cor-
egistration of angiographic images and echo images
into the workflow via ML-based probe detection and
automated registration updates, enabling clinical
teams to identify soft tissue–based structures that are
provided directly from the integrated ultrasound
system. TrueFusion can enable echocardiographers
and interventionalists to better communicate and
achieve more intuitive anatomical orientation during
challenging procedures. This may result in reduced
contrast usage, procedure time, and radiation expo-
sure (19). In the future, via DL, automated diagnosis
of imaging-based pathologies may be possible inde-
pendent of an imaging specialist.
CLINICAL DECISION SUPPORT. Clinical
support systems with cognitive computing are under
development and include self-learning systems using
ML, pattern recognition, and NLP to mimic human
thought processes (Figure 2). A multinational study,
CEREBRIA-1 (Machine Learning vs Expert Human
Opinion to Determine Physiologically Optimized
Coronary Revascularization Strategies), evaluated
whether an ML algorithm based on computational
interpretation of pressure-wire pull back data would
be similar to expert human interpretation for treat-
ment strategies in patients with stable coronary ar-
tery disease (20). The study included 1,008
instantaneous wave-free ratio pullback traces,
including 317 duplicates, which were analyzed by
both the ML algorithm and a multinational team of
interventionalists. They found that the computer-
based ML program was noninferior to the expert
consensus decision for both appropriateness of PCI
and determining PCI strategy (20).
decision IBM’s Watson for Health applies cognitive tech-
nology to extract and analyze information from the
EMR, lab reports, imaging reports, the published
medical reports, guidelines, and various Internet
sources (2,3). This technology combines ML and
1298 Sardar et al.
Artificial Intelligence and Interventional Cardiology
F I G U R E 2 Future Catheterization Laboratory With Artificial Intelligence-Enabled Technologies
Artificial intelligence-enabled future catheterization laboratory with clinical decision support system, voice-powered virtual assistant,
augmented reality platforms, and semiautonomous/autonomous robotic system.
systems neuroscience to build powerful, general-
purpose learning algorithms into neural networks
that mimic the human brain. The IBM Watson for
Oncology cognitive computing system can provide
confidence-ranked, evidence-based treatment rec-
ommendations for cancer. Treatment recommenda-
tions by Watson and a tumor board were concordant
in 96% of lung, 81% of colon, and 93% of rectal cancer
cases (21). Currently, IBM is developing Medical
Sieve, an automated cognitive assistant for cardiolo-
gists and radiologists designed to
decision-making. This IBM project has addressed
many modalities of cardiac imaging
automatic detection of coronary
angiography.
VIRTUAL REALITY, AUGMENTED REALITY,
AND AI
Although there are technological differences in vir-
tual reality (VR)/augmented reality (AR) and AI,
combined application of these technologies may be
useful for interventional procedures (Figure 2). VR
platforms are currently being used in periprocedural
planning of structural heart interventions, as well
as pre-procedure patient experiences to decrease
JACC: CARDIOVASCULAR INTERVENTIONS VOL. 12, NO. 14, 2019
JULY 22, 2019:1293–303
anxiety and stress (22). The FDA-approved True 3D
system developed by EchoPixel (Santa Clara, Califor-
nia) renders patient-specific anatomy in an intuitive,
interactive VR format (23). SentiAR Inc, a St. Louis,
Missouri–based company, received a $2.2 million
research grant from the National Institutes of Health
to develop an AR cardiac hologram technology that
allows real-time viewing, measurement and manip-
ulation of patient anatomy in a holographic display
for procedural guidance (24) (Figure 3). Similar
aid clinical AR systems can be used to overlay important infor-
mation required during procedures that is typically
including displayed on multiple monitors stacked around the
stenoses in interventionist.
VOICE-POWERED VIRTUAL ASSISTANTS. Voice-powered
virtual assistants, such as Apple’s Siri, Amazon’s
Alexa, and Google’s Assistant, employ AI speech
recognition that has now improved to the point of
exceeding human accuracy in transcribing conversa-
tion. Compared with typing on keyboards for
searching data in EMR or online medical information,
voice is easier, faster, more convenient, and with the
help of a unique voiceprint, can eliminate the need
for passwords (25). Voice-powered virtual assistants
use speech recognition and NLP to “understand”
and process spoken data for output. AI empowered,
JACC: CARDIOVASCULAR INTERVENTIONS VOL. 12, NO. 14, 2019
JULY 22, 2019:1293–303
voice-powered virtual assistants have the potential to
process the input of multimodal data/images, and
then present them in a meaningful way to the
physician or operator. A voice-powered virtual assis-
tant system in the catheterization laboratory could
help operators control equipment, navigate the EMR
system or access image libraries more effectively in a
hands-free fashion.
“BIG DATA” RESEARCH AND
PREDICTIVE ANALYSIS
“To date, big data, such as “omics” data, human gut
microbiome sequencing, social media, and cardiac
imaging, are too large and heterogeneous to be
stored, analyzed, and used to their full potential” (2).
AI techniques may solve this issue as well as allow for
automatic generation of new hypotheses, instead of
requiring that physicians postulate them (1,6).
Traditionally, regression-based statistical methods
are limited to using a small number of predictors,
which operate in the same way on everyone with
uniformity throughout their range. Moreover, in-
teractions mediating physiological processes may be
extremely complex to be captured using common
regression techniques. Unsupervised DL may facili-
tate exploration of novel factors in score systems and
better prediction analysis, or add hidden risk factors
to existing models (2,7). This could lead to new
models for antiplatelet/anticoagulant therapy,
bleeding versus stroke risk, mortality risk with pro-
cedures, and so on (2,6). ML-based techniques also
can highlight long-term outcomes or late complica-
tions for patients who have undergone a specific
procedure or are prescribed a specific drug. Previ-
ously, ML has been used in cardiology to predict
1-year mortality in heart failure patients and 5-year
mortality from coronary angiography datasets in pa-
tients with suspected coronary artery disease (4,7).
Several recent studies have used ML-based tech-
niques for advanced analysis of a dataset. Boone et al.
(26) evaluated the cardiovascular disease profile of
hematological malignancy patients undergoing PCI
and used bivariable analysis with bootstrapping
and result validation by an ML-generated neural
network. A study by Ganim et al. (27) assessed racial
disparities in procedural complications after trans-
catheter aortic valve replacement using ML-driven,
backward propagation neural networks. Topological
data analysis within an ML framework can be helpful
in large volume multidimensional data interpreta-
tion, such as describing new phenotypes of diseases
or combining imaging, clinical, and other nonstruc-
tured data.
Sardar et al. 1299
Artificial Intelligence and Interventional Cardiology
F I G U R E 3 Augmented Reality Cardiac Holographic Display for Procedural Guidance
INTEGRATION OF AI-BASED DECISION-MAKING
ACROSS ALL PHASES OF PATIENT CARE
In the future, interventional cardiologists will likely
see AI analysis of population- and patient-specific
data augmenting each phase of care. AI algorithms
will be used in the emergency department to triage
chest pain patients. In a pilot study, ML algorithms
showed 94% accuracy for predicting a myocardial
infarction in patients presenting with chest pain in
the emergency department (28). Automated analysis
of all pre-operative mobile and clinical data will
provide patient-specific risk scores for procedural
planning and yield valuable predictors to inform
post-procedural care. Algorithms may be applied to
patients with cardiogenic shock in order to deter-
mine who might benefit from mechanical circulatory
support. DL prediction models will predict the
periprocedural risk of death, bleeding, contrast
nephropathy, and stroke (1,2). Predictive analytics
with the use of cognitive computing may support
clinical decision-making and help prioritize tasks
in the catheterization laboratory. Intraprocedural
monitoring of various data may lead to the real-
time prediction and avoidance of adverse events.
Integration of pre-, intra-, and post-procedural
data could help to monitor recovery, predict com-
plications, and recommend the optimal duration
of medical therapy (2,5). After discharge, post-
operative data from personal devices could
continue to be integrated with data from the
hospitalization to optimize patient recovery and
decrease readmission rates. AI-enabled applications
and apps may encourage and incentivize healthier
behavior in individuals and may help with the
preventive strategies of a healthy lifestyle and
medication adherence (2).
1300 Sardar et al.
Artificial Intelligence and Interventional Cardiology
F I G U R E 4 Applications of Robotics in IC
Abbreviations as in Figure 2.
PHYSICAL APPLICATION OF AI
ROBOTICS. AI, through procedural automation, has
the potential to increase the value of robotics in the
catheterization lab by reducing the variability of
procedure time and improving overall patient care
(Figures 2 and 4). Although robotic interventions
show promise, limitations and hurdles remain. Cur-
rent systems do not comprehend the anatomy they
display to the interventionalist, the procedures they
are being used to perform, or know what the inter-
ventionalist intends to do. It is possible to enhance
these robotic tools using AI technologies such as
computer vision and image analysis (5). Although
truly autonomous robotic vascular procedures will
remain out of reach for some time, synergy across
fields will likely accelerate the capabilities of AI in
augmenting interventional care.
Corindus Vascular Robotics (Waltham, Massachu-
setts) recently received 510(k) clearance from the FDA
for the first automated robotic movement designed
for the CorPath GRX platform (29). The proprietary
software feature, named “Rotate on Retract,” is an
automated robotic movement that allows the oper-
ator to quickly navigate to a targeted lesion by auto-
matically rotating the guidewire upon joystick
retraction. Preclinical data demonstrated a significant
JACC: CARDIOVASCULAR INTERVENTIONS VOL. 12, NO. 14, 2019
JULY 22, 2019:1293–303
reduction in wiring time among a highly experienced
group of physicians when comparing robotic wiring
versus robotic wiring with Rotate on Retract enabled.
Future generations of vascular robotic platforms will
be more aware of the procedures being performed
and use that knowledge to provide intelligent assis-
tance to interventional cardiologists. Companies such
as Verb Surgical, a collaboration between Google and
Ethicon Endo-Surgery, have indicated that their sur-
gical robots will include ML and awareness, which
will aim to identify potential issues during a proced-
ure (30). They plan to link the robot to a cloud su-
percomputer service akin to IBM’s Watson, so that
information on thousands of similar procedures will
be accessible to both the surgeon and the robot to
improve performance (5,30). Additional projects un-
der development include the development of micro-
bots that can travel through blood vessels to deliver
medications to a specific target. Future applications
of such microbots include the potential to repair
damaged cells or perform microprocedures, which
might include a variety of vascular interventions (1,5).
Robots can be useful in interventional training
simulator and teleintervention. Outside the cathe-
terization lab, robots can help in post-procedural
rehabilitation, and can be useful in other areas of
health care such as pharmabotics, disinfectant robots,
supply chain robots, and personal assistance (2,5,31).
The true potential of AI in robotics remains to be seen
and is difficult to predict at this time. Therefore,
practicing interventional cardiologists should be
engaged in assessing the quality and applicability of
AI advances to ensure appropriate translation to the
clinical sector.
Several organizations and startup companies are
presently evaluating the application of AI-based
technologies in other areas of health care that may
influence IC. These include virtual nurses, digital
consultation, medication management by patients,
drug creation, health monitoring with wearable
health trackers and health care system analysis
(1,6,31).
LIMITATIONS OF AI-BASED TECHNOLOGIES
As with any novel technology, AI and its sub-
specialties are subjected to unrealistic expectations,
which may lead to disappointment and disillusion-
ment in the future (Figure 5). As noted by experts in
the field, “ML is a natural extension of traditional
approaches, not a magical device that can spin data
into gold” (32). There are instances where traditional
analytical methods can outperform ML or where the
JACC: CARDIOVASCULAR INTERVENTIONS VOL. 12, NO. 14, 2019
JULY 22, 2019:1293–303
F I G U R E 5 Advantages and Challenges of AI in IC
Abbreviations as in Figures 2 and 4.
addition of ML does not improve results (2,5). For
instance, Frizzell et al. (33) reported that a number of
ML algorithms did not improve prediction of 30-day
heart failure readmissions compared with more
traditional prediction models. “As more control is
ceded to algorithms, it is important to note that these
new algorithmic decision-making tools come with no
guarantees of fairness, equitability, or even veracity”
(32). Issues likely to arise with the application of ML/
DL on various datasets include: 1) issues with data
integrity (just a newer version of the classic adage:
“garbage in, garbage out”); 2) issues related to lack of
diversity in training datasets; and 3) an impaired or
absent ability to fully evaluate for methodological
bias in analysis. Of particular concern is the latter
issue as it relates to neural networks, which are based
on a ‘‘black box’’ design (1,5). Although the auto-
mated nature of neural networks allows for detection
of patterns missed by humans, human scientists are
left with little ability to assess how or why the com-
puter discerned such patterns. Human physicians,
therefore, must critically evaluate the predictions
generated by AI and interpret the data in clinically
meaningful ways (1,8). Large, well-curated datasets
are required for training of DL algorithms to provide
diagnostic and predictive capabilities. However, the
lack of large datasets of carefully annotated images
and videos have been limiting across various disci-
plines in medicine including IC. Interestingly,
Sardar et al. 1301
Artificial Intelligence and Interventional Cardiology
generative adversarial networks have been used to
compensate for this deficiency and synthetically
produce large image datasets, including angiograms
and echocardiograms, at high resolution that could be
used to help train deep neural networks (34).
There are concerns that robotization may lead to an
increase in unnecessary interventions, take focus
away from patient expectations, or exacerbate exist-
ing socioeconomic biases related to care delivery.
Wireless connectivity of wearable and implantable
devices, cloud-based AI technologies, and robots face
cyber security risks. A proposed solution to mitigate
this issue is a custom-built “health care blockchain”
technology that can store health information on an
encrypted digital ledger in order to minimize cyber
security risks (35).
One looming question is whether AI technologies
could eventually replace a human interventional
cardiologist. This does not look possible in the fore-
seeable future. Current technologies have several
shortcomings in comparison to activities of the hu-
man brain. Structurally, bio-inspired neural networks
resemble at best the outer layers of the retina or the
visual cortex where images are just sensed or repre-
sented (36). The capacity of complex decision making
or performing a procedure independently would
be quite arduous for current AI/ML algorithms. AI,
although widely publicized for its amazing perfor-
mance, is actually quite shallow in intelligence in its
1302 Sardar et al. JACC: CARDIOVASCULAR INTERVENTIONS VOL. 12, NO. 14, 2019

Artificial Intelligence and Interventional Cardiology JULY 22, 2019:1293–303

current form (1,36). Robotic systems geared towards
the technical performance of cardiovascular in-
terventions are far too primitive to operate indepen-
dently on patients. In the aviation industry, pilots
embraced autopilot technology; however, most pas-
sengers would not want to do away with pilots.
Similarly, we anticipate AI will assist rather than
replace the human operator in the catheterization
laboratory. The societal and ethical complexities of
applications of AI require further reflection, proof of
their medical utility, economic valuing, and devel-
opment of interdisciplinary strategies for their wider
application (37).
SYNERGIZING IC AND AI
AI, physicians should seek to expand involvement in
local, national, or international clinical data regis-
tries. As data cleaning techniques improve, registries
could become linked to expand their utility and in-
crease the availability of clinical, genomic, proteomic,
radiographic, and angiographic data available for AI-
based analysis. Interventionalists have the clinical
insight that can guide data scientists and engineers to
answer the right questions with the right data;
whereas engineers can provide automated, compu-
tational solutions to data analytics problems that
would otherwise be too costly or time-consuming for
manual methods. If appropriately developed and
implemented, AI has the potential to revolutionize
the way IC is taught and practiced.

There will always be a need for the expertise, CONCLUSIONS

compassion, and skill of a physician in the practice of
IC. AI is not a replacement for human intellect.
Rather, it has the potential to complement and rein-
force it. The current practice of IC already represents
a paradigm of clinician–machine interaction, and this
synergy is likely to accelerate in the coming age of AI.
Interventional cardiologists are uniquely positioned
to help drive these innovations rather than passively
wait for such technology to become useful. With AI-
based systems providing more information for pa-
tient care, interventional cardiologists may see their
roles shift. Physicians, including interventional car-
diologists, need to develop additional skills, which
machines cannot master in the near future, such as
ethics, leadership, and empathy. Training programs
must put a greater emphasis on how to optimize new
AI tools. Important issues include posing the right
questions to machines, interpreting their outputs,
identifying when machines make mistakes, and
integrating machine-generated data in patient care.
Because lack of data can limit the predictions made by
AI has initiated a paradigm shift in health care,
powered by the increasing availability of health care
data and rapid evolution of analytic techniques. It is
expanding its footprint in clinical systems, including
databases, image and intraprocedural video analysis,
evidence-based, real-time clinical decision support,
and robotics. The unique nature of interventional
practice leaves the interventionalist well-positioned
to help usher in the next phase of AI, one focused
on synergistic interaction between man and machine,
which ultimately will transform IC practice in efforts
to improve clinical care.
ADDRESS FOR CORRESPONDENCE: Dr. Jay Giri,
Hospital of the University of Pennsylvania,
Cardiovascular Medicine Division, Gates Pavilion,
9th Floor, 3400 Spruce Street, Philadelphia, Penn-
sylvania 19104. E-mail: giri.jay@gmail.com. Twitter:
@parthasardarmd, @jaygirimd.

REFERENCES

1. Shameer K, Johnson KW, Glicksberg BS,
Dudley JT, Sengupta PP. Machine learning in car-
diovascular medicine: are we there yet? Heart
2018;104:1156–64.
2. Krittanawong C, Zhang H, Wang Z, Aydar M,
Kitai T. Artificial intelligence in precision cardio-
vascular medicine. J Am Coll Cardiol 2017;69:
2657–64.
5. Hashimoto DA, Rosman G, Rus D, Meireles OR.
Artificial intelligence in surgery: promises and
perils. Ann Surg 2018;268:70–6.
6. Jiang F, Jiang Y, Zhi H, et al. Artificial intelli-
gence in healthcare: past, present and future.
Stroke Vasc Neurol 2017;2:230–43.
7. Johnson KW, Torres Soto J, Glicksberg BS, et al.
Artificial intelligence in cardiology. J Am Coll
large cohort of all-comers undergoing percuta-
neous coronary intervention: Comparison of five
contrast media. Int J Cardiol 2018;273:69–73.
10. Narula S, Shameer K, Salem Omar AM,
Dudley JT, Sengupta PP. Machine-learning algo-
rithms to automate morphological and functional
assessments in 2D echocardiography. J Am Coll
Cardiol 2016;68:2287–95.

3. Syeda-Mahmood T. Role of big data and ma-
chine learning in diagnostic decision support in
radiology. J Am Coll Radiol 2018;15 Pt B:569–76.
4. Henglin M, Stein G, Hushcha PV, Snoek J,
Wiltschko AB, Cheng S. Machine learning
approaches in cardiovascular imaging. Circ Car-
diovasc Imaging 2017;10:e005614.
Cardiol 2018;71:2668–79.
8. Al’Aref SJ, Anchouche K, Singh G, et al. Clinical
applications of machine learning in cardiovascular
disease and its relevance to cardiac imaging. Eur
Heart J 2019;40:1975–86.
9. Azzalini L, Vilca LM, Lombardo F, et al. Inci-
dence of contrast-induced acute kidney injury in a
11. Shah SJ, Katz DH, Selvaraj S, et al. Pheno-
mapping for novel classification of heart failure
with preserved ejection fraction. Circulation 2015;
131:269–79.
12. Small AM, Kiss DH, Zlatsin Y, et al. Text mining
applied to electronic cardiovascular procedure
reports to identify patients with trileaflet aortic
JACC: CARDIOVASCULAR INTERVENTIONS VOL. 12, NO. 14, 2019
JULY 22, 2019:1293–303
stenosis and coronary artery disease. J Biomed
Inform 2017;72:77–84.
13. Sohn S, Ye Z, Liu H, Chute CG, Kullo IJ. Iden-
tifying abdominal aortic aneurysm cases and con-
trols using natural language processing of
radiology reports. AMIA Jt Summits Transl Sci Proc
2013;2013:249–53.
14. Du T, Xie L, Liu X, et al. Intelligent recognition
of coronary angiography by deep learning tech-
nology: a novel computer-aided diagnostic sys-
tem. Paper presented at: Transcatheter
Cardiovascular Therapeutics Symposium (TCT) 18;
September 21–25, 2018; San Diego, CA.
15. Ciusdel C, Turcea A, Puiu A, et al. An artificial
intelligence based solution for fully automated
cardiac phase and end-diastolic frame detection
on coronary angiographies. Paper presented at:
Transcatheter Cardiovascular Therapeutics Sym-
posium (TCT) 18; September 21–25, 2018; San
Diego, CA.
16. Molony D, Hosseini H, Samady H. Deep IVUS: a
machine learning framework for fully automatic
IVUS segmentation. Paper presented at: Trans-
catheter Cardiovascular Therapeutics Symposium
(TCT) 18; September 21–25, 2018; San Diego, CA.
17. Fairbairn TA, Nieman K, Akasaka T, et al. Real-
world clinical utility and impact on clinical
decision-making of coronary computed tomogra-
phy angiography-derived fractional flow reserve:
lessons from the ADVANCE registry. Eur Heart J
2018;39:3701–11.
18. Fearon WF, Achenbach S, Engstrøm T. Accu-
racy of fractional flow reserve derived from coro-
nary angiography. Circulation 2019;139:477–84.
19. Siemens Healthineers. Siemens Healthineers
Receives FDA Clearance for TrueFusion Structural
Heart Disease Feature. September 6, 2017. Available
at: https://usa.healthcare.siemens.com/news/
fdaclearanceoftruefusionfeature.html. Accessed
October 19, 2018.
20. Davies J. CEREBRIA-1: machine learning vs
expert human opinion to determine physiologically
optimized coronary revascularization strategies.
Paper presented at: Transcatheter Cardiovascular
Therapeutics Symposium (TCT) 18; September 24,
2018; San Diego, CA.
21. Somashekhar SP, Sepúlveda MJ, Norden AD,
et al. Early experience with IBM Watson for
Oncology (WFO) cognitive computing system for
lung and colorectal cancer treatment (abstr).
J Clin Oncol 2017;35:15 Suppl:8527.
22. Project Brave Heart: Studying the Impact of
Virtual Reality Preparation and Relaxation Therapy.
Lucile Packard Children’s Hospital Stanford web-
site. Available at: http://www.stanfordchildrens.
org/en/innovation/virtual-reality/anxiety-research.
Accessed October 10, 2018.
23. Silva JNA, Southworth M, Raptis C, Silva J.
Emerging applications of virtual reality in cardio-
vascular medicine. J Am Coll Cardiol Basic Trans
Science 2018;3:420–30.
24. Improving Visualization and Interaction During
Transcatheter Ablation Using A Mixed Reality
System: First-In-Human Experience. SentiAR
Technology website. Available at: https://sentiar.
com/#!info-graph. Accessed October 10, 2018.
25. Steinhubl SR, Topol EJ. Now we’re talking:
bringing a voice to digital medicine. Lancet 2018;
392:627.
26. Boone D. Cardiovascular disease and percuta-
neous coronary intervention in hematological
malignancies: knowledge is power. Paper pre-
sented at: Transcatheter Cardiovascular Thera-
peutics Symposium (TCT) 18; September 21–25,
2018; San Diego, CA.
27. Ganim N, Monlezun D, Balan P. Racial dispar-
ities in procedural complications after trans-
catheter aortic valve replacement (TAVR): a
retrospective single-center case-control study.
Paper presented at: Transcatheter Cardiovascular
Therapeutics Symposium (TCT) 18; September
21–25, 2018; San Diego, CA.
28. Lindholm D, Holzmann M. Machine learning
for improved detection of myocardial infarction in
Sardar et al. 1303
Artificial Intelligence and Interventional Cardiology
patients presenting with chest pain in the emer-
gency department (abstr). J Am Coll Cardiol 2018;
71 Suppl:A225.
29. Corindus Receives FDA Clearance for First
Automated Robotic Movement in technIQ Series
for CorPath GRX Platform [press release]. March 5,
2018. Available at: https://goo.gl/Sb1B2i.
Accessed October 10, 2018.
30. Two Industry Leaders Make for One Extraor-
dinary Verb. Verb Surgical website. Available at:
http://www.verbsurgical.com/about/. Accessed
October 19, 2018.
31. Elson DS, Cleary K, Dupont P, Merrifield R,
Riviere C. Medical robotics. Ann Biomed Eng 2018;
46:1433–6.
32. Beam AL, Kohane IS. Big data and
machine learning in health care. JAMA 2018;319:
1317–8.
33. Frizzell JD, Liang L, Schulte PJ, et al. Predic-
tion of 30-day all-cause readmissions in patients
hospitalized for heart failure: comparison of ma-
chine learning and other statistical approaches.
JAMA Cardiol 2017;2:204–9.
34. Topol EJ. High-performance medicine: the
convergence of human and artificial intelligence.
Nat Med 2019;25:44–56.
35. Ichikawa D, Kashiyama M, Ueno T.
Tamper-Resistant Mobile Health Using Block-
chain Technology. JMIR Mhealth Uhealth 2017;5:
e111.
36. Sengupta PP, Adjeroh DA. Will artificial intel-
ligence replace the human echocardiographer?
Circulation 2018;138:1639–42.
37. He J, Baxter SL, Xu J, Xu J, Zhou X, Zhang K.
The practical implementation of artificial intelli-
gence technologies in medicine. Nat Med 2019;25:
30–6.
KEY WORDS artificial intelligence,
interventional cardiology