Acessibilidade / Reportar erro

Artificial Intelligence in Cardiology: Concepts, Tools and Challenges - “The Horse is the One Who Runs, You Must Be the Jockey”

Abstract

The recent advances at hardware level and the increasing requirement of personalization of care associated with the urgent needs of value creation for the patients has helped Artificial Intelligence (AI) to promote a significant paradigm shift in the most diverse areas of medical knowledge, particularly in Cardiology, for its ability to support decision-making and improve diagnostic and prognostic performance. In this context, the present work does a non-systematic review of the main papers published on AI in Cardiology, focusing on its main applications, potential impacts and challenges.

Artificial Intelligence/trends; Computer Systems/trends; Machine Learning/trends; Cardiovascular Diseases; Clinical Decision-Making

Resumo

Os recentes avanços ao nível de hardware e a crescente exigência de personalização dos cuidados associados às necessidades urgentes de criação de valor para os pacientes contribuíram para que a Inteligência Artificial (IA) promovesse uma mudança significativa de paradigma nas mais diversas áreas do conhecimento médico, em particular em Cardiologia, por sua capacidade de apoiar a tomada de decisões e melhorar o desempenho diagnóstico e prognóstico. Nesse contexto, o presente trabalho faz uma revisão não-sistemática dos principais trabalhos publicados sobre IA em Cardiologia, com foco em suas principais aplicações, possíveis impactos e desafios.

Inteligência Artificial/tendências; Sistemas de Computação/tendências; Aprendizado de Máquina/tendências; Doenças Cardiovasculares; Tomada de Decisão Clínica

Introduction

A person’s everyday life necessitates a huge amount of knowledge about the world and the volume of data in health grows exponentially throughout the world.11. Goodfellow I, Bengio Y, Courville A. Deep Learning. Cambridge: MIT Press; 2016. On the other hand, biomedical knowledge is always expanding in an active and dynamic way and cannot be processed or stored by a single human brain. This situation makes it very difficult for the contemporary physician to keep up-to-date with such a broad spectrum of new data and findings, as well as to use such information easily and in a timely manner.22. Dilsizian SE, Siegel, EL. Artificial intelligence in medicine and cardiac imaging: harnessing big data and advanced computing to provide personalized medical diagnosis and treatment. Curr Cardiol Rep. 2014;16(1):441. Adding to this framework are the significant burnout rates among health professionals33. Michel JB, Sangha DM, Erwin JP 3rd. Burnout among cardiologists. Am J Cardiol. 2017;119(6):938-40. , 44. Rotenstein LS, Torre M, Ramos MA, Rosales RC, Guille C, Sen S, et al. Prevalence of burnout among physicians: a systematic review. JAMA. 2018;320(11):1131-50. and the important impact of medical errors - which in the United States represent the third leading cause of death.55. Macary MA, Daniel M. Medical error – the third leading cause of death in the US. BMJ. 2016 May 3;353:i2139. This panorama brings with it the need to reorganize the productive structure of health services, associated with various challenges and new perspectives. Given that the current health system is generally unproductive and/or expensive, it is imperative to develop alternative and innovative strategies. The central focus for achieving this goal should be to increase the value for the patient – outcomes reached per dollar spent – so that good outcomes, efficiently obtained, are a target to be pursued.66. Porter ME. A strategy for health care reform – toward a value-based system. N Engl J Med. 2009;361(2):109-12.

Besides, the recent advances at hardware level related to parallel processing, the existence of several machine-learning methods and the huge amount of annotated data contributed for artificial intelligence (AI) to promote a significant paradigm shift in the most diverse areas of medical knowledge and, particularly in Cardiology, for its ability to support decision-making that can improve diagnostic and prognostic performance. These impacts ought to be evaluated from the perspective of patient safety, personalization of care, value creation for the patients, within a scope of technological surveillance – that gradually consolidates AI as fundamental for a medical practice of excellence.77. Dilsizian ME, Siegel EL. Machine meets biology: a primer on artificial intelligence in cardiology and cardiac imaging. Curr Cardiol Rep. 2018;20(12):139.

8. Johnson KW, Soto JT, Glicksberg BS, Shameer K, Miotto R, Ali M, et al. Artificial Intelligence in Cardiology. J Am Coll Cardiol. 2018;71(23):2668-79.

9. Krittanawong C, Zhang H, Wang Z, Aydar M, Kitai T. Artificial intelligence in precision cardiovascular medicine. J Am Coll Cardiol. 2017;69(21):2657-64.

10. Massalha S, Clarkin O, Thornhill R, Wells G, Chow BJW. Decision support tools, systems, and artificial intelligence in cardiac imaging. Can J Cardiol. 2018;34(7):827-38.
- 1111. Mesquita CT. Artificial intelligence and machine learning in cardiology – a change of paradigm. Int. J. Cardiovasc. Sci. 2017;30(3):187-8.

This scenario makes AI, given its importance, be considered by many as the new electricity. The main journals in cardiology have published reviews in this area and the number of articles on the subject follows a growing trend, as shown in Figure 1 – this behavior is also seen in other medical specialties, such as Neurology. Therefore, the present work performs a non-systematic review of the main papers published on AI in Cardiology, focusing on its main applications, potential impacts and challenges. The next section presents the conceptual fundamentals on the topic, followed by a discussion on why cardiology needs AI and its main tools. Finally, the main challenges, perspectives and conclusions are presented.

Figure 1
Evolution of the number of works relating (Artificial Intelligence or Machine Learning) and Cardiology. Source: Pubmed. Accessed on 12/15/2018. Mesh Words: Cardiology and Machine Learning.

What is artificial intelligence?

The term AI was used for the first time at the Dartmouth Conference in 1956.1212. Moore J. The Dartmouth College Artificial Intelligence Conference: The Next Fifty Years. AI Magazine. 2006;27(4):87-91. Nevertheless, the possibility of machines being able to simulate human behavior and actually think was raised earlier by Alan Turing in 1950, who developed a test in order to differentiate humans from machines – thus named Turing test.1313. Turing AM. Computing machinery and intelligence. Mind. 1950;59(236):433-60.

Basically, AI is the product of the combination of sophisticated mathematical models and computation, which allows the development of complex algorithms capable of emulating human intelligence. All this process starts with the construction of a database representative of the problem that one wishes to study – adequately collected and processed– called healthy data. This step is of fundamental importance, as the algorithms will probably not perform well if this prerequisite is not obtained: " garbage in, garbage out ".

The nature of these data is quite varied, ranging from socio-environmental, clinical-laboratory, omic-data (e.g., metabolome, proteome, epigenome, lipidome) to information on red, green and blue intensities (RGB system) of each pixel that composes an image, for example. Equally diversified sources of such data include those obtained from electronic medical records or even wearable devices. In this context, the term Big Data is used to describe a huge collection of data for which traditional methods of analysis are unsuccessful in analyzing, searching, interpreting and storing.99. Krittanawong C, Zhang H, Wang Z, Aydar M, Kitai T. Artificial intelligence in precision cardiovascular medicine. J Am Coll Cardiol. 2017;69(21):2657-64.

We highlight the use of these tools in problems of classification, regression, and clusterization. After obtaining healthy data and building the database, it is important to evaluate which mathematical models of AI are most appropriate for the problem that one wishes to solve. Then, the chosen models must be implemented using some programming language. A combination of models can also be useful. The results obtained by the algorithm should be analyzed in terms of both the coherence and suitability. These steps are summarized in Figure 2 .

Figure 2
Main Illustration.

Why does cardiology need artificial intelligence?

The development of AI algorithms has the advantage of not requiring many assumptions in relation to underlying data.88. Johnson KW, Soto JT, Glicksberg BS, Shameer K, Miotto R, Ali M, et al. Artificial Intelligence in Cardiology. J Am Coll Cardiol. 2018;71(23):2668-79. Another point is that the nature of these mathematical-computational models allows, from observational data, a high level of evidence due to its high performance, which certainly represents a significant paradigm shift in evidence-based medicine. It should be noted that traditional clinical trials are generally slow, expensive, time-consuming, and limited in size.1414. Price, WN. Big data and black-box medical algorithms. Sci Transl Med. 2018;10(471):pi:eaao5333. In addition, when the database is fed with more (healthy) data, in general, there is an improvement in the performance of the algorithms – which allows the studies to have a continuous character over time.

This new archetype can guide the allocation of scarce resources in the health area and facilitate the efficient and accurate identification of decisions that favor the individualization of care based on the flow of information that emerge from an integrated and complex ecosystem: it is a precision medicine.1515. Antman EM, Loscalzo J. Precision medicine in cardiology. Nat Rev Cardiol. 2016;13(10):591-602. , 1616. Price, WN II. Black Box medicine. Harvard J Law Tech. 2015;28(2):419-467. Therefore, it can be inferred that the practice of the cardiovascular sciences will have significant impacts, which will translate into a personalized approach and improved outcomes.

Basic concepts in artificial intelligence

A generic database can be arranged in a matrix of rows and columns. Each line denotes an element from a set of objects to be evaluated according to the same features. Each column, in turn, expresses the values of a given attribute for the various rows in the database and each line represents a lesson to be learned by the mathematical-computational model. In this way, the term Machine Learning (ML) brings with it a possibility of " learning " from a set of lessons. The term AI is often used interchangeably with the term ML. However, ML is a subset of AI algorithms related the ability of learn from a large amount of data. AI is wider and encompass performing tasks that are normally related to human intelligence such as pattern recognition, problem solving, understanding language or recognizing objects and sounds.1717. Dey D, Slomka PJ, Leeson P, Comaniciu D, Shrestha S, Sengupta PP, et al. Artificial intelligence in cardiovascular imaging: JACC State-of-the-Art review. J Am Coll Cardiol. 2019;73(11):1317-35.

It is often said that the types of learning can be:

  1. Supervised: when the algorithm receives information about each lesson as well as the labels associated with it, having an important role in relation to the prediction. For example, if it is desired to predict whether a patient is more susceptible to cough with the use of angiotensin-converting enzyme inhibitors, analysis should be performed based on a healthy database containing a group of patients that showed such a reaction and another group in which this fact was not observed.

  2. Unsupervised: when the lesson labels are not provided a priori , it is up to the algorithm to find hidden structures in the database. A hypothetical example is the clusterization of a database of patients with hypertrophic cardiomyopathy according to imaging findings.

  3. Reinforcement: inspired by behavioral biology, it is a kind of reward-based learning.1818. Shameer K, Johnson KW, Glicksberg BS, Dudley JT, Sengupta PP. Machine learning in cardiovascular medicine: are we there yet? Heart. 2018;104(14):1156-64. , 1919. Al’Aref SJ, Anchouche K, Singh G, Slomka PJ, Kolli KK, Kumar A, et al. Clinical applications of machine learning in cardiovascular disease and its relevance to cardiac imaging. Eur Heart J. 2018;40(24):1975-86.

Another important concept is that of cognitive computing. It can be understood as a set of self-learning systems intended to imitate the human thought process based on the use of ML tools, pattern recognition and natural language processing.99. Krittanawong C, Zhang H, Wang Z, Aydar M, Kitai T. Artificial intelligence in precision cardiovascular medicine. J Am Coll Cardiol. 2017;69(21):2657-64. IBM Watson is an example of cognitive computing in the medical field.2020. Somashekhar SP, Sepulveda MJ, Puglielli S, Norden AD, Shortliffe EH, Rohit Kumar C, et al. Watson for Oncology and breast cancer treatment recommendations: agreement with an expert multidisciplinary tumor board. Ann Oncol. 2018;29(2):418-23. , 2121. IBM. Watson. New York: IBM; 2019 [Cited in 2018 Oct 10]. Available from: https://www.ibm.com/watson/index.html.
https://www.ibm.com/watson/index.html...

Some artificial intelligence tools and applications

Currently, there is a multiplicity of models of ML each of them with diverse particularities, varied uses and limitations. The applications of some of these models in Cardiology are explained in the following paragraphs, while a brief description of each of them and their type is shown in Table 1 .

Table 1
– Brief description and classification of the main ML tools

  1. Support Vector Machine (SVM): used by Samad et al.,2222. Samad MD, Wehner GJ, Arbabshirani MR, Jing L, Powell AJ, Geva T, et al. Predicting deterioration of ventricular function in patients with repaired tetralogy of Fallot using machine learning. Eur Heart J Cardiovasc Imaging. 2018;19(7):730-8. to predict with success the deterioration of ventricular function in patients with repaired tetralogy of Fallot from a database of 153 patients with clinical, electrocardiographic and cardiac magnetic resonance imaging data. In relation to predicting any deterioration (minor or major) vs. no deterioration, the mean area under the curve (AUC) was 0.82 ± 0.06.2222. Samad MD, Wehner GJ, Arbabshirani MR, Jing L, Powell AJ, Geva T, et al. Predicting deterioration of ventricular function in patients with repaired tetralogy of Fallot using machine learning. Eur Heart J Cardiovasc Imaging. 2018;19(7):730-8. Berikol et al.2323. Berikol GB, Yildiz O, Özcan IT. Diagnosis of acute coronary syndrome with a support vector machine. J Med Syst. 2016;40(4):84. used clinical, laboratory (troponin I and CK-MB levels), ECG, and echocardiographic data from 228 patients who presented at the emergency department with chest pain for classification regarding the presence or absence of Acute Coronary Syndrome. Accuracy, sensitivity and specificity were, respectively, 99.19, 98.22 and 100%.2323. Berikol GB, Yildiz O, Özcan IT. Diagnosis of acute coronary syndrome with a support vector machine. J Med Syst. 2016;40(4):84. Betancur et al.2424. Betancur J, Rubeaux M, Fuchs TA, Otaki Y, Arnson Y, Slipczuk L, et al. Automatic valve plane localization in myocardial perfusion SPECT/CT by machine learning: Anatomic and clinical validation. J Nucl Med. 2016;58(6):961-7. also used SVM to more precisely define mitral valve plane (VP) positioning during left ventricular segmentation in Single-Photon Emission Computed Tomography (SPECT) exams. Images of 392 patients were analyzed and the good results obtained were compatible with the opinion of experts in the area – AUC: 0.82 [0.74-0.9] for regional detection of obstructive stenosis and ischemic total perfusion deficit areas.2424. Betancur J, Rubeaux M, Fuchs TA, Otaki Y, Arnson Y, Slipczuk L, et al. Automatic valve plane localization in myocardial perfusion SPECT/CT by machine learning: Anatomic and clinical validation. J Nucl Med. 2016;58(6):961-7.

  2. Naive Bayes (NB): Paredes et al.,2525. Paredes S, Rocha T, de Carvalho P, Henriques J, Morais J, Ferreira J. Integration of different risk assessment tools to improve stratification of patients with coronary artery disease. Med Biol Eng Comput. 2015;53(10):1069-83. used an NB fusion and genetic algorithm to predict the risk of occurrence of cardiovascular events (e.g., hospitalization or death) based on data from 559 Acute Coronary Syndrome-Non-ST Segment Myocardial Infarction (ACS-NSTEMI) patients. Sensitivity and specificity were, respectively, 79.8, 83.8.2525. Paredes S, Rocha T, de Carvalho P, Henriques J, Morais J, Ferreira J. Integration of different risk assessment tools to improve stratification of patients with coronary artery disease. Med Biol Eng Comput. 2015;53(10):1069-83.

  3. K-nearest neighbors (KNN): Al-Mallah et al.2626. Al-Mallah MH, Elshawi R, Ahmed AM, Qureshi WT, Brawner CA, Blaha MJ, et al. Using machine learning to the association between cardiorespiratory fitness and all-cause mortality (from the Henry Ford Exercise Testing Project). Am J Cardiol. 2017;120(11):2078-84. compared the prediction of all-cause mortality in 10 years between the classical logistic regression model and the KNN, considering a database of 34,212 patients with clinical information and information obtained after the treadmill test using the standard protocol of Bruce.2626. Al-Mallah MH, Elshawi R, Ahmed AM, Qureshi WT, Brawner CA, Blaha MJ, et al. Using machine learning to the association between cardiorespiratory fitness and all-cause mortality (from the Henry Ford Exercise Testing Project). Am J Cardiol. 2017;120(11):2078-84. The results obtained by this ML tool showed a sensitivity of 87.4% and specificity of 97.2%, better than the predictive performance of the traditional Atherosclerosis Cardiovascular Disease Risk Score (ASCVD).

  4. Genetic algorithms (GA): Smisek et al.2727. Smisek R, Hejc J, Ronzhina M, Nemcova A, Marsanova L, Kolarova J, et al. Multi-stage SVM approach for cardiac arrhythmias detection in short single-lead ECG recorded by a wearable device. Physiol Meas. 2018;39(9):094003. developed a wearable device to detect arrhythmias from the information record of a single-lead electrocardiogram. The data were analyzed from a combination of the (SVM), decision tree and threshold-based rules. Genetic algorithms were used to select the most appropriate characteristics to be used in the work. In relation to the detection of atrial fibrillation, an F1 score (harmonic mean of positive predictive value and sensitivity) of 0.81 was obtained.2727. Smisek R, Hejc J, Ronzhina M, Nemcova A, Marsanova L, Kolarova J, et al. Multi-stage SVM approach for cardiac arrhythmias detection in short single-lead ECG recorded by a wearable device. Physiol Meas. 2018;39(9):094003. Stuckey et al.,2828. Stuckey TD, Gammon RS, Goswami R, Depta JP, Steuter JA, Meine FJ 3rd, et al. Cardiac phase space tomography: a novel method of assessing coronary artery disease utilizing machine learning. PLoS One. 2018;13(8):e0198603. used the Cardiac Phase Space Tomography Analysis - a pioneering method that dispenses with the use of radiation and contrast, as well as performing exercises or pharmacological stress - combined with ML models (e.g., genetic algorithms) to analyze the thoracic phase signals. In this study, the authors used this tool to evaluate patients with coronary disease and chest pain who were referred by the physician for angiography. 606 patients were studied, and the results showed sensitivity of 92%, specificity of 62% and predictive value of 96% for coronary disease.2828. Stuckey TD, Gammon RS, Goswami R, Depta JP, Steuter JA, Meine FJ 3rd, et al. Cardiac phase space tomography: a novel method of assessing coronary artery disease utilizing machine learning. PLoS One. 2018;13(8):e0198603.

  5. Random Forests (RF): Samad et al.,2929. Samad MD, Ulloa A, Wehner GJ, Jing L, Hartzel D, Good CW, et al. Predicting survival from large echocardiography and electronic health record datasets: optimization with machine learning. JACC Cardiovasc Imaging. 2019;12(4):681-9. analyzed a database consisting of clinical and electrocardiographic variables to evaluate survival in 10 different periods of time (ranging from 6 to 60 months), considering a total of 171,510 patients. RF was used, with excellent results, better than those obtained through traditional scores such as the Framingham risk score and ACC/AHA guideline score. The area under the curve (AUC) was superior to 0.82.2929. Samad MD, Ulloa A, Wehner GJ, Jing L, Hartzel D, Good CW, et al. Predicting survival from large echocardiography and electronic health record datasets: optimization with machine learning. JACC Cardiovasc Imaging. 2019;12(4):681-9. Ambale-Venkatesh et al.3030. Ambale-Venkatesh B, Yang X, Wu CO, Liu K, Hundley WG, McClelland R, et al. Cardiovascular event prediction by machine learning: the multi-ethnic study of atherosclerosis. Circ Res.2017;121(9):1092-101. used information from noninvasive tests, questionnaires, biomarkers and imaging tests from 6,814 patients to construct 739 variables (features) in order to apply a variant of RF – called survivor random forests3131. Ishwaran H, Kogalur UB, Blackstone EH, Lauer MS. Random survival forests. Ann Appl Stat. 2008;2(3):841-60. – for predicting cardiovascular events (all-cause death, stroke, all cardiovascular disease, coronary heart disease, atrial fibrillation and heart failure), having performed better than established risk scores, e.g., MESA-CHD, AHA/ASCVD and Framingham, with increased prediction accuracy (decreased Brier score by 10%-25%).3030. Ambale-Venkatesh B, Yang X, Wu CO, Liu K, Hundley WG, McClelland R, et al. Cardiovascular event prediction by machine learning: the multi-ethnic study of atherosclerosis. Circ Res.2017;121(9):1092-101. , 3131. Ishwaran H, Kogalur UB, Blackstone EH, Lauer MS. Random survival forests. Ann Appl Stat. 2008;2(3):841-60.

  6. K-means: Cikes et al.3232. Cikes M, Sanchez-Martinez S, Claggett B, Duchateau N, Piella G, Butakoff C, et al. Machine learning-based phenogrouping in heart failure to identify responders to cardiac resynchronization therapy. Eur J Heart Fail. 2019;21(1):74-85. used a database consisting of clinical variables and echocardiographic parameters for which two models of ML, Kmeans and Multiple Kernel Learning were applied, in order to categorize the patients into mutually exclusive groups to evaluate the response to resynchronization therapy cardiac. A total of 1,106 patients were analyzed and four disjoint groups were identified, two of them with the best response to therapy.3232. Cikes M, Sanchez-Martinez S, Claggett B, Duchateau N, Piella G, Butakoff C, et al. Machine learning-based phenogrouping in heart failure to identify responders to cardiac resynchronization therapy. Eur J Heart Fail. 2019;21(1):74-85.

  7. Artificial Neural Networks (ANN): Kwon et al.,3333. Kwon JM, Lee Y, Lee Y, Lee S, Park J 2018;7(13):pii: e008678. in a multicenter study of 52,131 patients, constructed a deep learning-based early warning system capable of predicting the occurrence of cardiac arrest in a hospital. The model showed high performance when compared to traditional track-and-trigger systems. The area under the curve was 0.82.3333. Kwon JM, Lee Y, Lee Y, Lee S, Park J 2018;7(13):pii: e008678. Rubin et al.,3434. Rubin J, Parvaneh S, Rahman A, Conroy B, Babaeizadeh S. Densely connected convolutional networks for detection of atrial fibrillation from short single-lead ECG recordings. J Electrocardiol. 2018;51(6S):S18-S21. had promising preliminary results with the use of neural networks with convolutional architecture to evaluate electrocardiographic signs and to classify them in atrial fibrillation, sinus rhythm (normal) or noise – the F1 score achieved was 0.82.3434. Rubin J, Parvaneh S, Rahman A, Conroy B, Babaeizadeh S. Densely connected convolutional networks for detection of atrial fibrillation from short single-lead ECG recordings. J Electrocardiol. 2018;51(6S):S18-S21. Zhang et al.3535. Zhang J, Gajjala S, Agrawal P, Tison GH, Hallock LA, Beussink-Nelson L, et al. Fully automated echocardiogram interpretation in clinical practice. Circulation. 2018;138(16):1623-35. also used convolutional neural networks to analyze a database with 14,035 echocardiographic exams to detect the presence of diseases such as hypertrophic cardiomyopathy, cardiac amyloidosis and pulmonary arterial hypertension with a high performance: C statistics were respectively, 0.93, 0.87, and 0.85.3535. Zhang J, Gajjala S, Agrawal P, Tison GH, Hallock LA, Beussink-Nelson L, et al. Fully automated echocardiogram interpretation in clinical practice. Circulation. 2018;138(16):1623-35. Nakajima et al.3636. Nakajima K, Kudo T, Nakata T, Kiso K, Kasai T, Taniguchi Y, et al. Diagnostic accuracy of an artificial neural network compared with statistical quantitation of myocardial perfusion images: a Japanese multicenter study. Eur J Nucl Med Mol Imaging. 2017;44(13):2280-9. used an ANN to evaluate the presence of coronary disease after performing myocardial scintigraphy. Results were obtained with high accuracy and superior performance to the traditional scores used. For example, the AUC for patients with old myocardial infarction based on defects in rest stage was 0.97.3636. Nakajima K, Kudo T, Nakata T, Kiso K, Kasai T, Taniguchi Y, et al. Diagnostic accuracy of an artificial neural network compared with statistical quantitation of myocardial perfusion images: a Japanese multicenter study. Eur J Nucl Med Mol Imaging. 2017;44(13):2280-9.

  8. Gradient Boosting (GB): Mortazavi et al.3737. Mortazavi BJ, Bucholz EM, Desai NR, Huang C, Curtis JP, Masoudi FA, et al. Comparison of machine learning methods with national cardiovascular data registry models for prediction of risk of bleeding after percutaneous coronary intervention. JAMA Netw Open. 2019;2(7):e196835. used GB for prediction of risk of bleeding after percutaneous coronary intervention and demonstrated that these tools can help to identify patients who would benefit from strategies aiming to reduce the bleeding risk. A total of 3,316,465 procedures were analyzed and a C statistic of 0.82 was obtained.3737. Mortazavi BJ, Bucholz EM, Desai NR, Huang C, Curtis JP, Masoudi FA, et al. Comparison of machine learning methods with national cardiovascular data registry models for prediction of risk of bleeding after percutaneous coronary intervention. JAMA Netw Open. 2019;2(7):e196835. Hernesniemi et al.,3838. Hernesniemi JA, Mahdiani S, Tynkkynen JA, Lyytikäinen LP, Mishra PP, Lehtimäki T, et al. Extensive phenotype data and machine learning in prediction of mortality in acute coronary syndrome - the MADDEC study. Ann Med. 2019;51(2):156-63. also proposed a GB to predict mortality in acute coronary syndrome, analyzing 9,066 consecutive patients. The AUC was 0.89 and the model performed better than GRACE traditional score.3838. Hernesniemi JA, Mahdiani S, Tynkkynen JA, Lyytikäinen LP, Mishra PP, Lehtimäki T, et al. Extensive phenotype data and machine learning in prediction of mortality in acute coronary syndrome - the MADDEC study. Ann Med. 2019;51(2):156-63.

It is important to note that when using any ML model, one should keep in mind a major problem that may arise, called overfitting. It occurs when a model describes the examples very well (training subset) and performs poorly when applied to other instances of the same phenomenon.3939. Nannen V. The Paradox of Overfitting Volker Nannen. [thesis] . Países Baixos: Faculty of Artificial Intelligence at the University of Groningen; 2003. In addition, it is worth saying that there is no theoretical result that ensures that any of the AI algorithms is better than the others in any application. Thus, this choice depends on several variables, such as the nature of the problem under analysis, the time and resources available to solve the problem. The combination of techniques generating hybrid models can also be of great value. On the other hand, the use of tools for parallel processing, such as the Graphic Processing Unit (GPU), has been of great value in improving the performance of ML models, especially in relation to computational time needed to run them.

Challenges and future prospects

As previously highlighted, AI applications in cardiology have increased greatly in recent years and their growth potential is enormous. However, this scenario brings with it the need to overcome some challenges, such as: ethical limits of use (misuse), improvement of mathematical knowledge, acquisition of healthy data, development of security, need for collaboration, attention to errors and data-based care. All of this is discussed below and is summarized in Figure 2 .

  1. Challenge 1 – ethical limits of use (misuse): like all disruptive technology, the limits of ethics need to be rethought and widely discussed. ML algorithms can be misused and misleading. As an example, a work of great repercussion was published by Wang and Kosinski (2018). The authors used deep learning and obtained expressive results in the prediction of whether an individual is gay or not from a database of images of the study participants’ faces.5555. Wang Y, Kosinski M. Deep neural networks are more accurate than humans at detecting sexual orientation from facial images. J Pers Soc Psychol. 2018;114(2):246-57. Similarly, the same AI algorithms can be used to detect, for example, whether or not a patient will develop atrial fibrillation or any future cardiomyopathy. Could this information be used by companies to increase the amounts of their health plans or even deny membership to the plan due to a high cost? What if it is detected that a baby will be born with congenital heart disease due to the analysis of the genetic, clinical-laboratory and image (or other) data of its parents? This could open space for a kind of neoeugenia. This debate has gained an additional emphasis with the emergence of the CRISP-Cas9 technique, which allows DNA editing.5656. Ma H, Marti-Gutierrez N, Park SW, Wu J, Lee Y, Suzuki K, et al. Correction of a pathogenic gene mutation in human embryos. Nature. 2017;548(7668):413-9. In this context, by stimulating a debate with society on the subject, transparency and regulation are fundamental pillars to be preserved.

  2. Challenge 2improve math knowledge: the advent of this new kind of unbelievable human being ( Homo incredibile ), which supports its decisions in data, carries with it the fundamental role of mathematics and computation in this currently ongoing revolution. This revolution will bring unimaginable possibilities in medical practice, such as the construction of quality phenomappings - ML models developed with the aim of clustering patients in function of their large mass of phenotypic characteristics in order to facilitate the decision-making process.5757. Shah SJ, Katz DH, Deo RC. Phenotypic spectrum of heart failure with preserved ejection fraction. Heart Fail Clin. 2014;10(3):407-18. Thus, it is necessary that these competencies are stimulated early, mainly with a focus on solving problems related to the reality for which one wishes to promote improvements. This will certainly be reflected in a need to reformulate the cardiovascular contents (and why not say, medical content in general) of undergraduate and postgraduate courses in Medicine: a passive or merely expositive education, with an extensive load and that prioritizes the capacity of the student's memory seems, more and more, to be inadequate, as one realizes that Medicine must be a space for creativity and value generation.

  3. Challenge 3 – get healthy data: the use of healthy data is of fundamental value for the success of the algorithms. Thus, it is required that health units encourage their health professionals regarding the thoroughness at the level of data filling/obtaining as well as maintaining any data sources, from forms, electronic medical records, image data or even unconventional data, such as those obtained by Medina et al.5858. Medina EL, Mesquita CT, Loques Filho O. Healthcare social networks for patients with cardiovascular diseases and recommendation systems. Int J Cardiovasc Sci. 2016;29(1):80-5. - who developed a successful Online Social Networks Health tool in which the patient himself anonymously inserts health monitoring information, including physiological data, daily activities, emotional states, and interaction with others patients.5858. Medina EL, Mesquita CT, Loques Filho O. Healthcare social networks for patients with cardiovascular diseases and recommendation systems. Int J Cardiovasc Sci. 2016;29(1):80-5. Therefore, data management becomes as important as other routine behaviors in evidence-based medicine, such as proper handwashing or even the use of a defibrillator during cardiac arrest. In this way, the formation of multidisciplinary data teams and the constant training of the teams assume a primordial role. It is noteworthy that much of the slowness and difficulty that some health units have in using ML models is tied to absent or incipient healthy data.

  4. Challenge 4 – get security: the advent of these tools brings with it a fundamental concern with data security, to a level never before experienced, as access to such data by unauthorized persons can lead to catastrophic consequences for both health institutions and the patients. The creation of a security team plays an important role in this new process. The General Data Protection Regulation represents an advance in this direction. Blockchain and its variants are important tools that can improve security substantially.

  5. Challenge 5 – need for collaboration (the African proverb): there is an African proverb that says, " if you want to go fast go by yourself, but if you want to go far go with many ". This applies a lot to this data environment: collaboration between institutions allows the construction of huge healthy databases (Big Data), which tends to favor the performance of ML algorithms.

  6. Challenge 6 – dealing with errors: one important issue concerns the errors of AI models. It is inadequate to believe that such models are error-free. It may, for example, be the result of overfitting or occurring by using unhealthy data - which make the results unreliable. However, the practice has shown high performance in several applications. These models are probabilistic, and it is always desirable that their errors be minimal. This scenario has clinical implications, for example, an AI model that predicts with 99% probability that a patient has a greater propensity than the general population to have cardiac myocarditis or amyloidosis. There is a probability, although small, that this will not occur, and that the procedure adopted by the cardiologist is inadequate. In that case, the question is who can be held accountable in these cases? Is it appropriate? Should the patient sign a consent form in these cases? Certainly, the solution includes robust regulation of the use of these tools and strengthening of a new type of relationship: physician-patient-data.

  7. Challenge 7 – data-based care management: while ML's tools follow an inexorable path, on the other hand, several healthcare professionals remain fearful about these tools because of its possible ability to replace physicians in their tasks. However, when the history of Medicine is remembered, it is worth mentioning, for example, that the appearance of automated machines to perform the whole blood count did not replace the hematologist, but rather resulted in a greater speed of the work process and allowed the professional to be able to act in other important issues in the specialty.

The central idea is to provide better support for decision-making, including better performance. It is data-driven care management with a high dynamism and constant updating - which will promote greater personalization of care5959. Bittencourt MS. From evidence-based medicine to precision health: using data to personalize care. Arq Bras Cardiol. 2018;111(6):762-3. and a real-time evaluation of the experience of the health system users, aiming at generating value for the patient. In this context, the mechanical tasks will be substitutable and a diversity of new tasks will be included into the routine of the cardiologist of precision, from the adequate construction of the databases to the critical reflection on the results obtained by the mathematical-computational models, as well as the development of an adequate physician-patient-data relationship. Therefore, there is a migration of human skills as well as the expansion of their capabilities from the emergence of new tools, which should be part of the technical arsenal of the 21stcentury cardiologist. This panorama allows us to compare ML models to a horse and doctors to jockeys: " the horse is the one who runs , you must be the jockey ".

Conclusions

AI, in fact, has been shown to be a fundamental tool for the clinical practice of current cardiology. Several applications have been successfully performed and have allowed significant improvements from a diagnostic and therapeutic point of view and in relation to personalized care. To be able to use such tools, it is imperative that healthy data be used, which certainly implies a new design in the modus operandi of many health services. The nature of these data is varied and includes new sources, such as wearable devices and omic-data. On the other hand, this new digital ecosystem requires an acquisition of knowledge not traditionally found in regular medical courses. Therefore, a curricular redesign is required and ought to be object of a profound debate and specific actions.

On the other hand, the entire panacea brought by AI is not free from challenges such as: the ethics limits of its use, the necessity of improving math knowledge, the building of an ecosystem that ensure high levels of security and confidentiality for the patients, the acquisition of healthy data, the needs of expand the physician-patient-data association, the necessity of collaboration and the data-based care management. In this context, the cardiologist-jockey (or physicians in general) must be a protagonist of changes and has to replace an eventual fear of the tools by a greater involvement with the objective of generating value for the care. It is important to keep in mind possible challenges and obstacles to be overcome and to maintain an engagement and critical sense in the search for solutions: "the horse is the one who runs, you must be the jockey".

Referências

  • 1
    Goodfellow I, Bengio Y, Courville A. Deep Learning. Cambridge: MIT Press; 2016.
  • 2
    Dilsizian SE, Siegel, EL. Artificial intelligence in medicine and cardiac imaging: harnessing big data and advanced computing to provide personalized medical diagnosis and treatment. Curr Cardiol Rep. 2014;16(1):441.
  • 3
    Michel JB, Sangha DM, Erwin JP 3rd. Burnout among cardiologists. Am J Cardiol. 2017;119(6):938-40.
  • 4
    Rotenstein LS, Torre M, Ramos MA, Rosales RC, Guille C, Sen S, et al. Prevalence of burnout among physicians: a systematic review. JAMA. 2018;320(11):1131-50.
  • 5
    Macary MA, Daniel M. Medical error – the third leading cause of death in the US. BMJ. 2016 May 3;353:i2139.
  • 6
    Porter ME. A strategy for health care reform – toward a value-based system. N Engl J Med. 2009;361(2):109-12.
  • 7
    Dilsizian ME, Siegel EL. Machine meets biology: a primer on artificial intelligence in cardiology and cardiac imaging. Curr Cardiol Rep. 2018;20(12):139.
  • 8
    Johnson KW, Soto JT, Glicksberg BS, Shameer K, Miotto R, Ali M, et al. Artificial Intelligence in Cardiology. J Am Coll Cardiol. 2018;71(23):2668-79.
  • 9
    Krittanawong C, Zhang H, Wang Z, Aydar M, Kitai T. Artificial intelligence in precision cardiovascular medicine. J Am Coll Cardiol. 2017;69(21):2657-64.
  • 10
    Massalha S, Clarkin O, Thornhill R, Wells G, Chow BJW. Decision support tools, systems, and artificial intelligence in cardiac imaging. Can J Cardiol. 2018;34(7):827-38.
  • 11
    Mesquita CT. Artificial intelligence and machine learning in cardiology – a change of paradigm. Int. J. Cardiovasc. Sci. 2017;30(3):187-8.
  • 12
    Moore J. The Dartmouth College Artificial Intelligence Conference: The Next Fifty Years. AI Magazine. 2006;27(4):87-91.
  • 13
    Turing AM. Computing machinery and intelligence. Mind. 1950;59(236):433-60.
  • 14
    Price, WN. Big data and black-box medical algorithms. Sci Transl Med. 2018;10(471):pi:eaao5333.
  • 15
    Antman EM, Loscalzo J. Precision medicine in cardiology. Nat Rev Cardiol. 2016;13(10):591-602.
  • 16
    Price, WN II. Black Box medicine. Harvard J Law Tech. 2015;28(2):419-467.
  • 17
    Dey D, Slomka PJ, Leeson P, Comaniciu D, Shrestha S, Sengupta PP, et al. Artificial intelligence in cardiovascular imaging: JACC State-of-the-Art review. J Am Coll Cardiol. 2019;73(11):1317-35.
  • 18
    Shameer K, Johnson KW, Glicksberg BS, Dudley JT, Sengupta PP. Machine learning in cardiovascular medicine: are we there yet? Heart. 2018;104(14):1156-64.
  • 19
    Al’Aref SJ, Anchouche K, Singh G, Slomka PJ, Kolli KK, Kumar A, et al. Clinical applications of machine learning in cardiovascular disease and its relevance to cardiac imaging. Eur Heart J. 2018;40(24):1975-86.
  • 20
    Somashekhar SP, Sepulveda MJ, Puglielli S, Norden AD, Shortliffe EH, Rohit Kumar C, et al. Watson for Oncology and breast cancer treatment recommendations: agreement with an expert multidisciplinary tumor board. Ann Oncol. 2018;29(2):418-23.
  • 21
    IBM. Watson. New York: IBM; 2019 [Cited in 2018 Oct 10]. Available from: https://www.ibm.com/watson/index.html
    » https://www.ibm.com/watson/index.html
  • 22
    Samad MD, Wehner GJ, Arbabshirani MR, Jing L, Powell AJ, Geva T, et al. Predicting deterioration of ventricular function in patients with repaired tetralogy of Fallot using machine learning. Eur Heart J Cardiovasc Imaging. 2018;19(7):730-8.
  • 23
    Berikol GB, Yildiz O, Özcan IT. Diagnosis of acute coronary syndrome with a support vector machine. J Med Syst. 2016;40(4):84.
  • 24
    Betancur J, Rubeaux M, Fuchs TA, Otaki Y, Arnson Y, Slipczuk L, et al. Automatic valve plane localization in myocardial perfusion SPECT/CT by machine learning: Anatomic and clinical validation. J Nucl Med. 2016;58(6):961-7.
  • 25
    Paredes S, Rocha T, de Carvalho P, Henriques J, Morais J, Ferreira J. Integration of different risk assessment tools to improve stratification of patients with coronary artery disease. Med Biol Eng Comput. 2015;53(10):1069-83.
  • 26
    Al-Mallah MH, Elshawi R, Ahmed AM, Qureshi WT, Brawner CA, Blaha MJ, et al. Using machine learning to the association between cardiorespiratory fitness and all-cause mortality (from the Henry Ford Exercise Testing Project). Am J Cardiol. 2017;120(11):2078-84.
  • 27
    Smisek R, Hejc J, Ronzhina M, Nemcova A, Marsanova L, Kolarova J, et al. Multi-stage SVM approach for cardiac arrhythmias detection in short single-lead ECG recorded by a wearable device. Physiol Meas. 2018;39(9):094003.
  • 28
    Stuckey TD, Gammon RS, Goswami R, Depta JP, Steuter JA, Meine FJ 3rd, et al. Cardiac phase space tomography: a novel method of assessing coronary artery disease utilizing machine learning. PLoS One. 2018;13(8):e0198603.
  • 29
    Samad MD, Ulloa A, Wehner GJ, Jing L, Hartzel D, Good CW, et al. Predicting survival from large echocardiography and electronic health record datasets: optimization with machine learning. JACC Cardiovasc Imaging. 2019;12(4):681-9.
  • 30
    Ambale-Venkatesh B, Yang X, Wu CO, Liu K, Hundley WG, McClelland R, et al. Cardiovascular event prediction by machine learning: the multi-ethnic study of atherosclerosis. Circ Res.2017;121(9):1092-101.
  • 31
    Ishwaran H, Kogalur UB, Blackstone EH, Lauer MS. Random survival forests. Ann Appl Stat. 2008;2(3):841-60.
  • 32
    Cikes M, Sanchez-Martinez S, Claggett B, Duchateau N, Piella G, Butakoff C, et al. Machine learning-based phenogrouping in heart failure to identify responders to cardiac resynchronization therapy. Eur J Heart Fail. 2019;21(1):74-85.
  • 33
    Kwon JM, Lee Y, Lee Y, Lee S, Park J 2018;7(13):pii: e008678.
  • 34
    Rubin J, Parvaneh S, Rahman A, Conroy B, Babaeizadeh S. Densely connected convolutional networks for detection of atrial fibrillation from short single-lead ECG recordings. J Electrocardiol. 2018;51(6S):S18-S21.
  • 35
    Zhang J, Gajjala S, Agrawal P, Tison GH, Hallock LA, Beussink-Nelson L, et al. Fully automated echocardiogram interpretation in clinical practice. Circulation. 2018;138(16):1623-35.
  • 36
    Nakajima K, Kudo T, Nakata T, Kiso K, Kasai T, Taniguchi Y, et al. Diagnostic accuracy of an artificial neural network compared with statistical quantitation of myocardial perfusion images: a Japanese multicenter study. Eur J Nucl Med Mol Imaging. 2017;44(13):2280-9.
  • 37
    Mortazavi BJ, Bucholz EM, Desai NR, Huang C, Curtis JP, Masoudi FA, et al. Comparison of machine learning methods with national cardiovascular data registry models for prediction of risk of bleeding after percutaneous coronary intervention. JAMA Netw Open. 2019;2(7):e196835.
  • 38
    Hernesniemi JA, Mahdiani S, Tynkkynen JA, Lyytikäinen LP, Mishra PP, Lehtimäki T, et al. Extensive phenotype data and machine learning in prediction of mortality in acute coronary syndrome - the MADDEC study. Ann Med. 2019;51(2):156-63.
  • 39
    Nannen V. The Paradox of Overfitting Volker Nannen. [thesis] . Países Baixos: Faculty of Artificial Intelligence at the University of Groningen; 2003.
  • 40
    Cortes C, Vapnik V. Support-vector networks. Mach Learn. 1995;20:273-97.
  • 41
    Bayes T. An essay towards solving a problem in the doctrine of chances. By the late Rev. Mr. Bayes, F. R. S. communicated by Mr. Price, in a letter to John Canton, A. M. F. R. S. Philos Trans R Soc Lond. 1763;53:370-418.
  • 42
    Webb GI, Boughton JR, Wang Z. Not so naive bayes: aggregating one-dependence estimators. Mach Learn. 2005;58(1):5-24.
  • 43
    Watkins, DS. Fundamentals of matrix computations. 2th ed. New York: Wiley-Interscience; 2002.
  • 44
    Cover T, Hart P. (1967). Nearest neighbor pattern classification. IEEE Trans Inf Theory. 1967;13(1):21-7.
  • 45
    Fix, E., Hodges, J.L. Discriminatory analysis, nonparametric discrimination: Consistency properties. Technical Report 4, USAF School of Aviation Medicine, Randolph Field, Texas, 1951.
  • 46
    Holland JH. Adaptation in natural and artificial systems. 2th ed. Cambridge, MA: MIT Press; 1992.
  • 47
    Breiman L. Random forests. Mach Learn. 2001;45(1):5-32.
  • 48
    Ho TK. Random decision forests. In: Proceedings of the 3rd International Conference on Document Analysis and Recognition; 1995 Aug 14-16; Montreal. Washington, DC: IEEE Computer Society; 1995. p.278-82.
  • 49
    MacQueen JB. Some Methods for classification and Analysis of Multivariate Observations. Proc. Fifth Berkeley Symp. on Math. Statist. and Prob. 1967;1:281-97.
  • 50
    McCulloch WS, Pitts W. A logical calculus of ideas immanent in nervous activity. Bull Math Biophys. 1943;5(4):115-33.
  • 51
    Rosenblatt F. The perceptron: a probabilistic model for information storage and organization in the brain. Psychol Rev. 1958;65(6):386-408.
  • 52
    Broomhead DS, Lowe D. Multivariable functional interpolation and adaptive networks. Complex Syst. 1988;2:321-55.
  • 53
    Friedman JH. Greedy function approximation: a gradient boosting machine. Ann Stat. 2001;29(5):1189-232.
  • 54
    Friedman JH. Stochastic gradient boosting. Comput. Stat. Data Anal. 2002;38(4):367-78.
  • 55
    Wang Y, Kosinski M. Deep neural networks are more accurate than humans at detecting sexual orientation from facial images. J Pers Soc Psychol. 2018;114(2):246-57.
  • 56
    Ma H, Marti-Gutierrez N, Park SW, Wu J, Lee Y, Suzuki K, et al. Correction of a pathogenic gene mutation in human embryos. Nature. 2017;548(7668):413-9.
  • 57
    Shah SJ, Katz DH, Deo RC. Phenotypic spectrum of heart failure with preserved ejection fraction. Heart Fail Clin. 2014;10(3):407-18.
  • 58
    Medina EL, Mesquita CT, Loques Filho O. Healthcare social networks for patients with cardiovascular diseases and recommendation systems. Int J Cardiovasc Sci. 2016;29(1):80-5.
  • 59
    Bittencourt MS. From evidence-based medicine to precision health: using data to personalize care. Arq Bras Cardiol. 2018;111(6):762-3.
  • Study Association
    This article is part of the thesis of Doctoral submitted by Erito Marques de Souza Filho, from Universidade Federal Fluminense.
  • Ethics approval and consent to participate
    This article does not contain any studies with human participants or animals performed by any of the authors.
  • Sources of Funding
    There were no external funding sources for this study.

Publication Dates

  • Publication in this collection
    29 May 2020
  • Date of issue
    Apr 2020Apr 2020

History

  • Received
    21 Dec 2018
  • Reviewed
    16 July 2019
  • Accepted
    28 Aug 2019
Sociedade Brasileira de Cardiologia - SBC Avenida Marechal Câmara, 160, sala: 330, Centro, CEP: 20020-907, (21) 3478-2700 - Rio de Janeiro - RJ - Brazil, Fax: +55 21 3478-2770 - São Paulo - SP - Brazil
E-mail: revista@cardiol.br