John J. Bergan
1
famously quoted—In vascular surgery, no change for the better has occurred that wise
and good men have not opposed. Such change is inevitable.
Historically, the first revolution in vascular surgery was by Alexis Carrel,
2
who won the Nobel Prize in physiology/medicine in 1912 for suturing of blood vessels
and transplantation of organs. However, it took almost 45 years for the second vascular
revolution to occur, which involved 2 of the world’s most outstanding cardiovascular
surgeons—Dr Michael E. DeBakey and Dr Denton A. Cooley.
3
Both pioneered cardiovascular operations at the Baylor College of Medicine and Methodist
Hospital in Houston in the 1950s that have stood the test of time. During that time,
Charles Theodore Dotter, the father of intervention radiology, accidentally performed
a vascular dilatation in 1963, followed by intentional percutaneous angioplasty by
a balloon in 1964, to set the stage for the third vascular revolution of angioplasty
and the role of interventional radiology.
4
In September 1990, the fourth vascular revolution was brewing in Argentina, as Dr
Parodi, helped by Julio Palmaz and Hector Barone, performed the first reported endovascular
aortic aneurysm repair (EVAR) under epidural anesthesia.
4
The human capabilities that are most critical to success—the only ones that help your
organization become more resilient, more creative and more, well, awesome—are precisely
the ones that can’t be “managed”.—Whitehurst and Gary Hamel.
5
Vascular evolution has shown that change is imminent, largely driven by inefficiencies
in the provision of healthcare services, rising costs and growing levels of litigation.
Technologies that improve productivity and enhance safety will become an integral
part of healthcare delivery in modern medicine. However, many of us are yet reluctant
to encompass a shift in technology and step beyond the comfort of traditional practice.
According to the World Economic Forum report, one-third of essential skillsets in
2025 will comprise technological skills not yet considered imperative to the job today.
6
Evolution in technology will shape the future of medicine. Based on this trend, we
can predict that the cardiovascular global workforce will undergo a massive technological
shift leading to the “Fifth Cardiovascular Revolution.”
Knowledge and proficiency in technology will be an integral part of the skillsets
required for clinical practice, and beyond doubt, artificial intelligence (AI), machine
learning (ML), and Big Data analysis (BDA) will transform global healthcare practice.
The evolution of these technologies will have profound implications on our workforce
beyond anyone’s comprehension.
6,7
Fewer organizations will fill these critical roles with skillsets that will evolve
and emerge from the Fifth Cardiovascular Revolution.
Artificial Intelligence (AI), Machine Learning (ML), Deep Learning (DL), Natural Language
Processing (NLP), and Big Data Analysis (BDA)
Artificial intelligence is the exhibition of human intelligence through machines required
for problem-solving.
8
Machine learning is the subset of AI that allows computers to learn and automate through
computational algorithms without human intervention. In this context, DL is a subset
of ML that enables machines to train and learn through an artificial neural network.
Similarly, NLP is the branch of AI that helps computers understand human language,
whereas BDA involves the analysis and extraction of information from a large volume
of data obtained through various sources.
Advances in computing power, data storage capacities and hardware acceleration allowed
the storage of large volumes of medical data in hospitals and clinical data centers.
This presents an excellent opportunity to exploit AI and BDA as well as the technological
advances within ML, DL, and NLP fields to facilitate, improve, and support medical
decisions. This will allow us to exploit the diversity of medical and healthcare big
data to perform complicated tasks with precision and accuracy.
9
The digitization and availability of these data sets have encouraged many researchers
to explore various ML, DL, and NLP techniques to support healthcare applications.
Among the applications in healthcare is to help and assist in medical diagnosis and
outcome prediction. However, there are endless opportunities for AI and BDA to support,
improve, and even revolutionize the healthcare industry. These fast-growing technologies
can enhance diagnostics and care delivery, improve patient–doctor communication, and
smoothen hospital administration and community health management with AI-powered healthcare
solutions. Big data from diverse sources and medical systems can be fed to AI algorithms,
for example, to build interfaces for medical software and devices to create intelligent
applications for optical character recognition.
Currently, the major beneficiaries of the 21st century advances in big data, ML, and
data science are global industries and financial sectors because they have the resources
to hire the necessary staff to transform their products and acquire the knowledge
to be ahead of the game. However, the algorithms developed for these industries offer
considerable opportunities to further clinical care and medical research.
Implications of AI in Cardiovascular Medicine
As cardiovascular specialists and key opinion leaders, we must prepare to meet the
fast-growing technological demands and beyond. We must develop, deliver and learn
how to survive the technology dominating future as innovators and early adaptors,
and lead through the “Fifth Cardiovascular Revolution.”
10,11
Artificial intelligenceI, ML, and BDA in healthcare could drive real value and transform
care delivery for patients and caregivers, raising the bar for a new gold standard
that empowers patients to experience personalized and frictionless healthcare journeys.
The application of AI and BDA to a cardiovascular health strategy offer limitless
opportunities as cardiovascular disease is the commonest cause of morbidity and mortality
worldwide. The current results promise to provide a continuous source of data-driven
insights to optimize biomedical research, precision imaging, early public health primary
intervention, and overall superior healthcare quality improvement.
12
One major goal in the application of AI and BDA is to accurately predict cardiovascular
morbidity and mortality over time. One promising avenue to explore; is to employ AI
and BDA to identify patients at high-risk for cardiovascular medical emergencies;
in particular, AI-powered technologies could empower us to accurately predict when
the disease will occur, and once the disease is evident, we can predict when a patient
will relapse or transit into another disease state. Other promising applications include
the integration of image and language processing so that images of various pathologies
can be mapped and radiomics can be used to consolidate visual and textual data to
understand relevant disease features. This can allow for more precise predictions.
There is a growing area of research in “Multimodal Data Analysis,” which is focused
on the integration and interpretation of diverse data of various types and modalities.
13
This allows the AI techniques and their accompanying technologies of ML, DL, and NLP
to learn the relationship between many complex data, such as demographics, risk factors,
genomics, proteomics, and radiomics.
14
Healthcare systems have a lot to learn from other industries that employ best practices
such as “Six Sigma” to build robust AI algorithmic solutions to enhance efficiency
and provide seamless workflows.
15
Through AI and BDA applications, the potential exists to reinforce our expertise in
supervised and unsupervised cardiovascular practices with AI and BDA skills and knowledge,
to understand the use of the state-of-the-art ML, neural networks, and DL approaches,
and to build intelligent systems that make better decisions with the minimum human
help possible.
An exciting avenue that we strongly recommend is employing supervised ML and DL techniques
and algorithms to develop taxonomies from data sets containing multiple heterogeneous
inputs.
Supervised ML and a data set containing multiple heterogeneous inputs require 3 levels
of algorithms. The first algorithms accurately manage the “high sparsity data sets”
and fill the missing entry points by properly using its regularization function. The
second algorithm is a mathematical high-performance “kernel trick” as a support vector
machine that derives accurate predictions in situations where the relationship between
features and the outcome is nonlinear. Finally, the third algorithm is an artificial
neural network with complex architecture and heavily modifiable parameters that led
to the widespread use of high-definition clinical imaging and video recognition in
many challenging applications.
14
Integrating AI and BDA into the cardiovascular field will allow medical staff to devote
more time to caring for patients rather than doing other tasks that AI algorithms
can execute with precision. Such tasks include (1) checking results continuously,
(2) alarming physicians with changes in baselines, and (3) following agreed protocols
for emergency intervention. This will allow for neutralizing human error and creating
a robust, experienced system that can handle any catastrophe at any given time with
accuracy. Continuous monitoring systems that employ AI algorithms to recognize abnormalities
in patients’ clinical status offer great potential to ensure timely and life-saving
intervention. Various features are used in such algorithms, including patients’ clinical
signs and symptoms, medical laboratory investigations, high-definition imaging modalities,
and haemodynamic and clinical data activity feeds.
16
The field of radiomics offers fascinating potential for cardiovascular diagnostics.
It is a promising approach for characterizing imaging through morphological and functional
imaging data to enhance diagnostics and treatment. Radiomics could bring order out
of chaos by translating data sources that are spatially and temporally heterogeneous
to organized data through ML algorithms.
17,18
The European Society of Radiology (ESR)
19
performed a survey to determine the radiologists’ position toward AI and BDA technologies.
The survey exposed the shocking results that radiologists do not consider the implementation
of AI systems into radiological diagnostics as a probable field for AI application
in the next 10 years.
19
Half of the radiologists foresaw a negative impact of AI systems on future job opportunities
with fear that technological advancements could lead to a nearly two-thirds decrease
in their workload. However, what those surveys failed to realize is that AI and BDA
reduce time working on diagnostics and thereby free up time to provide stronger interactions
and increased communication with other clinicians. It is expected that radiologists
who adopt AI and BDA technologies will replace radiologists who do not.
20
On a positive note, AI-based algorithms can detect major cardiovascular diseases in
60% of asymptomatic subjects.
21,22
Moreover, such algorithms can be an alternative to conventional image processing methodologies
for image segmentation, registration, classification, and enhancement, making DL-based
approaches the best tactics for computer vision and medical image processing.
21,22
These successful examples allow us to move forward from theory to the clinical cardiovascular
practice of ML for clinicians and medical researchers. The basic principle is to validate
such complex tasks, emphasizing NLP and image recognition.
14
It is essential to understand that with time, the application of AI, intelligent swarming,
and knowledge-centered services (KCSs) will grow exponentially in cardiovascular medicine,
and the significant economic burden associated with its initial implementation will
decrease not only over time but also the return on investment will be worth the initial
capital buy-in (Figure 1).
Figure 1.
Predicted rise in the application of AI, ML, BDA, and related technologies in cardiovascular
medicine with a decrement in the application cost over time. AI, artificial intelligence;
ML, machine learning; BDA, Big Data analysis.
Intelligent Swarming
Intelligent swarming is the concept of using the required resources to solve the problems
without escalations to a higher level (Figure 2).
23
Figure 2.
New model swarming with the proposed collaboration-based process during intelligent
swarming.
Intelligent swarming supports a collaborative work environment and prevents significant
time-lapses, providing speedy healthcare delivery. For example, to improve collaboration
with the complex issues of trauma patients, as an example, from level 1 trauma to
level 3 trauma, health organizational silos have resulted in competition and scapegoating
between departments (Figure 3). The support issues are complex, and level 1 trauma-support
members often cannot solve problems in isolation. With a lack of communication and
failure to share data, patient care can be inappropriately escalated within or between
emergency departments. This is a clear example of where risks to patients can arise,
whereby clinical signs get missed, unwarranted transfers are enacted, and lives are
lost. Intelligent swarming has the potential to solve these issues.
Figure 3.
Traditional escalation process that is time-consuming and labor intensive.
If we employ intelligent swarming to address healthcare delivery issues, we can save
precious time and reduce costs by bypassing traditional hierarchical and obstructive
healthcare delivery systems.
Healthcare providers have been insidiously considering intelligent swarming or “collective
intelligence” concepts for many years. However, its ultimate implementation is dependent
on the members’ willingness and courage to try something that has not been done before,
as the elements of the framework are still in a discovery phase.
24,25
Implementing intelligent swarming includes the power of collective thinking and collective
experience.
Knowledge-Centered Service (KCS)
A KCS is about treating knowledge as power and a business asset.
26
Early benefits of KCS are improved resolution times and first call resolution, reduced
escalations, improved employee skills, job satisfaction, and confidence. These benefits
translate into less stress and potentially better retention rates, dramatic improvement
in self-service success, and reduced training time. Similarly, long-term benefits
include identification of user behaviors and trends, improved location of resolutions
via AI-driven solutions, and better analytical predictive and pre-emptive abilities
that enable proactive clinical engagement. The successful implementation of KCS enables
organizations to improve their ability to capitalize on their collective knowledge
and wisdom.
A KCS increases efficiency as team members spend less time searching for information
and more on helping patients and maintaining a sound health administration. A visible
store of knowledge will identify early that something has gone off base, predict adverse
events before they occur and inform how to correct them quickly. It makes the unknown
known.
Knowledge-centered service opposes the traditional knowledge engineering approach,
which was designed to capture the minimal data necessary to permit the timely transmission
of knowledge to many people. The 5 principles of KCS include—(1) create abundance
(share and learn more), (2) evolve and collective experience (fine-tune the work tasks
and increase the knowledge base), (3) demand-driven and reuse (knowledge is a by-product
of interaction and double-loop feedback), (4) resolve (engage all task holders and
empower the smart), and (5) improve and reward learning (to motivate all comers).
Knowledge-centered service is a many-to-many model through extensive data analysis,
deep ML, and AI algorithms (Figure 4).
Figure 4.
Principle of KCS. KCS, knowledge-centered services.
Limitations and Opportunities
The best way to predict the future is to invent it—Alan Key.
27
There is a lack of basic AI knowledge or an understanding of AI principles and terminology.
Furthermore, there is a paucity of skills and confidence in the use and application
of AI solutions.
28
Therefore, we have a stronger need for formalized education on AI to prepare the current
and prospective workforce for the upcoming clinical integration of AI in healthcare
to safely and efficiently navigate a digital future. As innovative and early adaptors,
we must concentrate on the needs of early learners depending on their age, sex, and
highest qualification to ensure optimal integration.
Transforming workforce productivity is an important area where AI delivers excellent
value through automation. There is plenty of hype, jargon, and abstract technology,
making it difficult for nonexperts to identify the most exciting opportunities to
apply AI into everyday practice. Learning the basics of AI and its associated technologies,
such as ML, DL, NLP, and BDA, begins with the terminology’s proper understanding.
Undoubtedly, the future will dictate that the knowledge of these technologies is essential
and that their applications will be crucial for healthcare practitioners.
Artificial intelligence–associated technologies will result in the computerization
of jobs with a substantial displacement of the human workforce. A Deloitte collaboration
with the Oxford Martin Institute predicted that AI could push 35% of U.K. jobs out
of existence over the next 10 years.
29,30
During the COVID-19 pandemic, working from home imposed enormous challenges on the
medical sectors, where many of us operated virtually to avoid the risk of spreading
COVID. As a result, this created specific new challenges such as ensuring reliable
internet connections and high-quality video calls. Thus, there is a pressing need
to solve such challenging problems by automating around 60% of all interactions by
2024 using AI-powered systems and self-services (Figure 5).
31,32
Figure 5.
Increasing applicability of artificial intelligence in cardiovascular medicine —25%
in 2018 to 60% in 2024.
The application of AI to healthcare will not be seamless. Artificial intelligence
researchers face many challenges during the design and development of AI-, ML-, or
DL-based algorithms, especially in healthcare. Among these challenges are data set
shift, accidental fitting of confounders, unintended discriminatory bias, the challenges
of generalization to new populations, and the inadvertent negative consequences of
new algorithms on health outcomes.
31
–33
These challenges could be neutralized by employing and adapting performance metrics
that are intuitive to clinicians and ideally go beyond measures of technical accuracy
to include quality of care and patient-driven/oriented outcomes. Timely and safe AI
translation into clinically validated and robust regulated systems will benefit everyone.
Regulators must balance the pace of innovation with the potential for harm alongside
thoughtful postmarket surveillance. This ensures that patients are not endangered
by any AI interventions, nor are they denied access to advantageous innovations.
31
–34
The purpose of AI is not about replacing humans’ tasks with machines; it is about
changing the routine work usually carried out by humans who can focus on more purposeful
work and optimize workplace productivity.
One of the existing challenges in rendering AI algorithms friendly to clinical practice
is the lack of harmonized accessible data. Most healthcare data are not readily available
for ML algorithms. Data are often buried in several medical imaging archival systems,
pathology systems, electronic health records (EHRs), prescribing tools and insurance
databases, which are challenging to bring together.
35
–37
Ethical Implications
We are at the kindergarten stage of the “Fifth revolution.” Considering the medico-legal
implications of AI outcomes, there is potential for critical and controversial issues
that are not yet clearly defined. Artificial intelligence–based system accountability
represents a challenge that will require robust regulations.
38
In the past, humans have made healthcare assessments singularly. Although intelligent
machines provide collective decision-making, it also raises issues like accountability,
transparency, permission, and privacy.
Many AI algorithms lack transparency, particularly for DL and ML-based algorithms
such as those used for image analysis, as it is practically impossible to interpret
or explain. If a patient is informed that an image has led to an impending medical
problem, the patient will more than likely want to know why. Deep learning algorithms,
and even physicians who are generally unfamiliar with their operation, may be unable
to provide any explanation.
Moreover, future patients will likely receive medical information from AI-based systems
that they would prefer to receive from an empathetic clinician face to face.
39
Robust clinical governance must act responsibly and establish governance mechanisms
to limit negative implications. Artificial intelligence and BDA are potent tools,
and their powerful technologies will impact human societies. It will require continuous
attention and thoughtful policy for many years to come.
40
We must better understand the complex and evolving relationship between clinicians
and human-centered AI tools in an evolving clinical environment.
41
–45
It is still challenging to provide diagnosis and treatment recommendations through
AI-based systems; however, we expect AI will ultimately master that domain. Given
the rapid advances in image analysis, it is most likely to impact the field of medicine—radiology
and pathology first, leading to images being solely examined by a machine.
We can foresee that AI and BDA technologies will extensively influence our cardiovascular
clinical practice within 5 years and augment patient care efforts. Over time, human
clinicians will gear toward job plans and tasks that allow for uniquely human skills
like empathy, persuasion, and integration of the bigger picture.
40
Healthcare providers who will not join the fifth cardiovascular revolution by embracing
AI programs will ultimately lose their jobs over time.
Artificial intelligence and BDA will improve scarce healthcare resources with personalized
precision patient management plans. It will inform and impact policies and guidelines.
It will also expedite clinical trials.
46
Furthermore, AI will create a paradigm shift in “disease management,” from providers
waiting for individuals to become sick and present with symptoms and signs to healthcare
organizations increasingly focused on disease prevention by proactively monitoring
healthy individuals, performing preventative and wellness interventions, and managing
prevention and wellness for at-risk individuals.
47
–49
Current medical education is antiquated; AI and BDA will pioneer a new division in
medicine. This professional hybrid physician, part-computer engineer and data scientist,
will increasingly embark on new roles and shift from “oracle” to “counsellor” to the
“COMENDA” physiCian, cOMputer ENgineer, DAta scientist. In this AI and BDA—empowered
workforce, the proficiency to decipher clinical problems together will become more
important than the knowledge of each practitioner.
50
–52
Salient Point
Deciphering AI/ML/BDA will grant clinicians and researchers without previous experience
the ability to critically scrutinize these techniques. It is mandatory to understand
the way algorithms are created, as medical practitioners will rely heavily on such
contemporary technology, which might not always perform as expected. Currently, models
are unable to achieve perfect performance. The legend of the Google Flu Trends model
failure offers stark scrutiny of consequences mitigated by an inability to understand
AI/ML/BDA models, which was employed and implemented superficially to improve Google
health trends.
53
Machine language algorithms training is mandatory to abolish and lessen the risks
of entrenching biases in predictive algorithms as medical practice bias-based risks
have been constantly acknowledged. If left unchecked, it jeopardizes the ethical use
of data-driven automation.
54
Conclusion
Artificial intelligence/ML/BDA momentum will transform and disrupt how medicine works.
Thus far, the enthusiasm has not been met by the ease of access to clinician training
targeted to the knowledge and skillsets required of any medical practitioner to deal
with such disruptive technology.
It is prudent to equalize the need for AI/ML/BDA plans that actively manages and lessens
potential accidental consequences while not conceding to marketing hype and profit
motives.
Artificial intelligence–, ML-, and DL-based approaches are the “payback” for the investment
in the implementation of EHRs. Electronic health records have provided support for
extensive data collection that every clinician, patient, and family would want, but
exploiting these data for healthcare provision is impossible without computer base
assistance. Artificial intelligence and BDA will empower our healthcare systems to
monitor patients remotely and intercept emergent clinical scenarios. The creation
of the AI-empowered healthcare specialist is crucial for a brighter future.