Keywords

1.1 Introduction

Artificial Intelligence (AI) is perhaps the greatest technological revolution of the 21st century. Its beginnings can be placed in the late 50s, when the famous English mathematician Alan Turing began to wonder what conditions are needed to consider a machine as intelligent. His ideas were reflected in his seminal article “Computing Machinery and Intelligence” [1]. However, the main concept was coined in 1956 by John McCarthy in the Dartmouth Summer Research Project on Artificial Intelligence [2], defining AI as “the science and engineering of making intelligent machines, especially intelligent computer programs”. Currently, practically no technological development escapes the use or influence of AI. We can find it in everyday situations such as the coffee machine that prepares us a cappuccino or sophisticated ones such as those that allow us to calculate the most optimal path so that an emergency vehicle (ambulance, fire truck, etc.) finds the optimal route to save lives or even very complex ones like a robot performing a complex surgery on a patient with a critical heart condition. In that sense, the IA has two main branches: virtual and physical. The first one is related with the informatics approaches where the algorithms are used for creating complex computer systems. The second one is close to robotics, where the algorithms are used to give the correct movement to the physic parts of a devices to perform an specific task.

But, what is intelligence? A very simple question with a no simple answer. In fact, we don’t have a general agreement on the definition of intelligence, but some good approaches have been presented during the last century, such as the capacity to give good answers from the point of view of truth or facts [3], the ability to continue an abstract thinking [4], the combination of abilities necessaries to survive in a particular culture [5], the ability to solve problems, to create products valued based on cultural settings [6] and from Random House Unabridged Dictionary (2006), the “capacity for learning, reasoning, understanding, and similar forms of mental activity; aptitude in grasping truths, relationships, facts, meanings, etc.”. It seems the learning capacity is an important issue when we try to explain what intelligence is.

Can a machine be intelligent as a human being? For now and hopefully for the future, the answer is no. But a very good approach is to give the machines the capacity of learning from data originated in a source, for example environmental data for a good prediction about future conditions. In that sense, if during the learning process the machine gets enough capacity to predict future scenarios, and making good predictions for taking correct decisions, we will be in front of an intelligent machine. But the reasoning and understanding capacities are not present in that machine.

For creating an intelligent machine, the learning process is essential because it is where the machine can identify patterns from the data and define rules on how to use them to make its best prediction regarding a future situation. However, data is not always required to make a machine perform tasks as an intelligent being would. Who has not ever played chess with the computer? Normally chess programming is based on rules of piece movements, being unnecessary data regarding the moves of other players. But these rules intrinsically contain a series of patterns that the programmer has programmed into algorithms. Clearly, the algorithm that represents the cybernetic player can be improved by finding new patterns that the programmer never imagined. This is where once again the massive processing of data from other players’ plays becomes extremely relevant. Indeed, the algorithm that represents the cybernetic player can be continuously improved until it reaches the point where it learns from patterns of so many plays that it practically covers a large part of the universe of possibilities that a human player would have, making the machine beat the man, which has already happened, but it has not been a process of reasoning, but rather of processing capacity.

There is no doubt that the applications developed and to be developed based on AI are beyond any prediction that we can make at the moment. However, it is practically an assumed fact that the world we know is undergoing dizzying changes where AI in many cases has taken a leading role. This is how in healthcare and in medicine in general, the next step in its development cannot be taken without the support of AI.

Since immemorial time, medicine has been focused on the cure of diseases and ailments of all kinds of patients [7], who have in common that symptoms have first manifested before decisions can be made, which in many cases may arrive late. However, the focus is changing, where the maintenance of healthy begins to to be relevant. Indeed, medicine is moving from a reactive to a proactive discipline being participatory, predictive, personalized and preventive P4 medicine [8]. The foregoing will be possible to the extent that we can obtain information and knowledge of the data related to the health of patients, in particular their DNA, electronic medical records and the environment in which they have lived, among other data. This is a huge challenge, starting with the maintenance and consolidation of data that ensures secure access for study purposes and of course to help patients with their ailments. Followed by the data storage and processing capacity for the extraction of patterns that help all health personnel in tasks such as diagnosis, treatment, medication, etc. And finally, to provide adequate help to both the patient and the health personnel to achieve the primary objective that goes beyond the cure of the disease, we refer to the maintenance of good health in the population. Given this scenario, artificial intelligence is playing a key role in massive data processing that allows the creation of countless applications, which cover practically all branches of medicine, such as radiology, nephrology, nutrition, psychiatry, neurology, etc. [9].

This chapter is organized as follow. Section 1.2 introduces the process about how the machines learn. In Sect. 1.3, the possible applications of artificial intelligence in medicine is shown. A short review about papers related with P4 medicine is developed in Sect. 1.4. And finally, as chapter’s summary, some reflections on the current and future potential of the application of artificial intelligence in medicine are shown.

1.2 How Do Machines Learn?

How do we human beings learn? through examples which are reflected in our neural network that then allow us to analyze sounds, images, movements, structures, tastes, etc. In this sense, the automatic learning of a machine is very similar, that is, many examples are needed so that the machine can extract patterns from the data that then allows it to extrapolate a future decision. When they receive data that did not belong to the set of examples as input and for which you are looking for an output, it will give you the best prediction based on what it had learnt before [10].

According to [11] an algorithm is capable to learn when given a task, training experience, and a performance measure, so its achievement on the task improves with experience. Based on experiences learned, the algorithms can take decisions, make predictions, classify objects, predict cancer in a patient, give directions to the robots, recognize the human voice, and a huge set the possible uses and applications. In this context, the data play a central role in achieving a correct training of the algorithms. In other words, if the data are not clean and consolidated, the training of the algorithm will be affected and its learning capacity will be reduced, which will affect all the uses described above, making it impossible to create applications.

Machine learning is the concept that groups the set of algorithms that are capable of learning from the data. It was defined by Arthur L. Samuel as a “field of study that gives computers the ability to learn without being explicitly programmed” [12]. Depending of type of data and the algorithm to use, the learning process can be classified in supervised, unsupervised, reinforcement [13], and deep learning [14].

In supervised learning, a set of training data is created where the input and output vector are known. Then the algorithm learns the associations between inputs and outputs to create a mathematical model, which is validated and tested with real data before it can be used. A different situation happens in unsupervised learning where the algorithm learns the associations in the training data but the output is unknown, identifying undiscovered patterns. In reinforcement learning the algorithm learns a behaviour through trial-and-error to maximize a defined reward. Finally, deep learning allow to process big amount of raw data to extract patterns for detection or classification of issues.

1.2.1 Machine Learning Process

Giving a collection of data we are trying to extract patterns and rules about how to use them. The basic assumption is that the data was not created randomly, i.e., there are intrinsic relationships among them, for example regularities, which can then be detected and represented as patterns that allow a machine learning algorithm to be trained. In other words, if the data are dirty, their dispersion is high or there are no possible relationships among them, then there will be no machine learning algorithm able to create an analysis, predict, classify, or model the phenomenon under study. In that sense, statistical techniques are used to analyze the quality of the data and make prospects of its possible uses in a training process of a machine learning algorithm [15].

Figure 1.1 shows a conceptual diagram about the machine learning algorithms, where the deep learning one is part of the artificial neural network family. Depending on the kind of data to be processed and the phenomenon under study to be modeled, is the machine learning algorithms to use.

Fig. 1.1
A block diagram depicts the machine learning algorithms. Deep learning includes is part of an artificial neural network.

Machine learning algorithms, concepts and classes, based on [14]

Full size image

Figure 1.2 shows the general learning process for machine learning algorithms focused in supervised learning. From a set of row data, the target data, i.e. the data to be considered to analyze the phenomenon under study, are selected. Next, a feature vector transformation process is developed. The feature vectors set is divided in three parts: training, validation, and testing. The training data are used to train the model by applying a selected machine learning algorithm. Then the model is validated to check if there are problems of overfitting/underfitting. And finally, testing is applied to get some effectiveness measure such as accuracy. After the whole process, the model is put into production, being useful for taking decisions.

Fig. 1.2
The process chart explains the training workflow. The steps include row data, data target, complete set of feature vectors, model, and finally production depiction on computer.

Machine learning general training process workflow

Full size image

1.2.2 Machine Learning in Medicine

Considering the current amount of data related to people’s health, and that its future growth will be exponential, it is no longer possible by using traditional methods of extracting information and knowledge, fundamentally manual, can obtain any benefit from these data [16]. In this way that machine learning algorithms have become the new way of processing large volumes of data, in many cases complex and multivariable, and in particular medical data [17].

Machine learning algorithms and a subset of them, like deep learning, allow to create a computational model to process medical data for getting novel patterns from raw data, for instance, digitized X-ray images of a human chest to detect if the patient has pneumonia or COVID-19 [18]. The learning model is trained by using a set of learning techniques, in this case an object detection technique to predict if a new chest X-ray image is classify as normal, pneumonia, or COVID-19.

Medical data are available in an enormous variety of types and sources, which include diseases, public health data, prognosis, diagnosis, images, time series, etc., and everything indicates that its growth will continue to be exponential, as more and better sensor devices are added to medical exams [19]. Almost by construction, medical data are complex to analyze, even when there are very few records to analyze a disease, the mere fact that these records have high dimentionality makes them a challenge in machine learning. Additionally, in many cases it is not entirely clear what methods have been used to ensure the quality of medical data, which directly influences the predictive quality of the model to be trained [20].

There is no doubt that the applications of machine learning, and deep learning in particular, in medicine are enormous and will continue to expand in the incoming years. In [17] presents a study on the use of machine learning in medical research in recent years. From there it follows that the most used algorithms have been Random Forrest, Decision Tree, Support Vector Machine, Naïve Bayes, Artificial Neural Networks, and clustering techniques. Especial attention with Deep Learning algorithms, used in an important set of papers, in particular Convolutional neural network (CNN) for medical imagine processing [21].

1.3 Artificial Intelligence in Medicine

During the last decade, Artificial Intelligence research has grown exponentially in projects related to medicine. Indeed only in 2016, AI projects related to medicine attracted much more investment than AI projects related to other sectors of the global economy [22, 23]. What factors explain this heightened growth and interest?, among the reasons we can find is the great capacity of AI to improve the efficiency and effectiveness of healthcare delivery. AI in medicine aims to optimize clinical decision-making by using data related with health to provide information and patterns for improving the healthcare of the individuals.

Then, what AI algorithms can do for the future of medicine? It is an inescapable fact that in the developed world, the amount of data related to each person’s health is growing by leaps and bounds. Also that the reality of the quality and consolidation of these data in many countries is doubtful. However, steps are being taken in the right direction to achieve the digitalization of computer systems related to electronic medical data, continuous improvement in data quality, interoperability of systems, and security protocols that guarantee that the access, processing, and use of the data will be carried out in accordance with the law [24]. In this context, there is only one question: what do we do with medical data? This is where AI plays a central role. With the avalanche of data that health information systems are producing day by day, the number of exams that patients must perform, DNA analyzes that are less expensive every day, and the data provided by the patients themselves through applications for Smartphones or other devices, the possibility of a human being able to analyze all this is simply impossible, which is why advanced mechanisms are required in data processing and in the presentation of information and patterns for making relevant clinical decisions.

Consider the following situation. A person suffers from a severe headache and tells a friend about his ailment. Immediately the friend takes out a medicine that he carries in his backpack that a doctor diagnosed him with years ago and that has always helped him with his headaches. Immediately afterwards, he gives a pill to his friend so that he can make the strong pain that he is suffering disappear. Several questions immediately arise: (1) if they are different persons, why give them the same drug? (2) does the friend know the person’s medical history?; (3) The dose delivered is correct?; (4) Will there be any contraindications? And we could continue asking questions, but the fact is that the advisable thing would be that if they are different persons, the treatment at least would not have to be the same. This is the essence of precision medicine that is the basis of the medicine of the future. Precision medicine, sometimes called personalized medicine is defined by the National Institute of Health (NIH) precision medicine as “refers to a new treatment and prevention method based on understanding of individual gene, environment and life-style” [25]. And to achieve this, information and knowledge extracted from quality data per individual are required, considering their medical history, DNA, environment, and lifestyle.

Figure 1.3 shows the integration of several data sources related with the patients, and the respective data processing by using artificial intelligence tools, machine learning algorithms and deep learning algorithms. The outcome aims to provide information, patterns, knowledge, etc, to improve the decision-making for different users, such as healthcare professionals and patients.

Fig. 1.3
The block diagram depicts artificial intelligence as an enabling factor for future medicine.

Artificial intelligence as enable factor for future medicine

Full size image

In Fig. 1.3, the collection of data presented is only a part of the data that are related to the life and healthcare of people. Among these data we also find those from online social networks [26] and those provided by people through smartphone applications related to healthcare [27]. Clearly, the more quality data that can be collected from an individual, the better the diagnosis to cure a disease or the advice to maintain their health, making the patient central in their healthcare. However, such amount of data cannot be processed directly by human beings, so a large data processing capacity is needed together with advanced algorithms for the extraction of information, patterns, and knowledge. The final result points to multiple users, starting with the patient, being able to make correct use of this new knowledge, which is key in healthcare.

There is no doubt that AI is causing a great positive revolution in terms of people’s healthcare. However, not everything is a rose garden. As it is sensitive personal data, its processing must be carried out under strict security measures and ethical considerations [28]. Indeed, what would happen when analyzing the DNA data of an individual, it is detected that she/he has a high probability of developing diabetes in the future, will this be a prevalence? will health insurance cover this disease? will the individual suffer some kind of discrimination? We must not forget that the use of AI in healthcare is to help people in an ethical and responsible manner [29], taking care that their personal data is not used for purposes other than to support health maintenance or cure the disease that affects an individual. In no way can they be used for other purposes that violate the fundamental human rights [30].

1.4 AI Applications in Medicine

As mentioned above, the development of high-throughput, data-intensive research assays, and technologies in medicine has created a need for researchers to develop strategies to analyze, integrate, and interpret the massive amounts of data they generate. Traditional approaches do not allow working with ’big data’, which presents the opportunity to improve people’s health, so the use of AI techniques suggest that they could be particularly appropriate. In addition, the results of the application of medical exams reveal a great heterogeneity in the pathophysiological factors and processes that contribute to a disease, suggesting that there is a need to adapt or ’customize’ the medicines to the nuanced characteristics and, often, unique to individual patients [31].

P4 Medicine represents the foundations of a model of clinical medicine, which offers the opportunity to modify the healthcare paradigm: the participation of the individual is essential to put into practice the other three aspects of P4 Medicine with each patient [32].

In Fig. 1.4 a summary about the P4 Medicine is shown. The future of health services focuses on giving people a complete picture of the many factors that affect their health. Personalized medicine proposes the integration of numerous biological data points, including molecular, cellular, and phenotypic longitudinal measurements, as well as individual genome sequences, to better define the health or well-being of each person, predict transitions to disease and guide medical interventions [33,34,35]. The implementation of the P4 from the clinical point of view will create predictive and personalized models that represent the well-being of each patient or of a disease, which will allow the design of new pharmacological tests that take into account the heterogeneity of responses to therapies and disease stratification [36].

Fig. 1.4
A circle diagram depicts the participatory, predictive, personalized, and preventive components of medicine.

Summary scheme of the definitions and components of the 4Ps of medicine: predictive, preventive, personalized, and participatory. Own elaboration based on [37]

Full size image

1.4.1 Predictive Medicine

Predictive medicine is a branch of medicine that aims to identify patients at risk of developing a disease, allowing for early prevention or treatment of that disease. Single or more commonly multiple assays are used to identify markers of future disease disposition [38].

The goal of predictive medicine is to obtain and catalog the characteristics of individual patients, analyze that data to predict an individual patient’s risk for an outcome of interest, predict which treatment in which individual will be most effective, and then intervene before the event occurs. However, previously limited to the realm of genetics, genetics-based risk prediction has been of limited benefit because whole genome sequencing is still relatively expensive, and most diseases affecting today’s population depend on different variables [39]. These are not only influenced by genetics but also by demographic, psychosocial data, eating habits, stress levels, etc. Therefore, the need arises to collect and organize different types of data to perform a more complete analysis of people.

In recent years, machine learning algorithms such as deep learning that can account for complex interactions among features is shown to be promising in predictive modeling in healthcare [40]. Artificial intelligence can help doctors make better clinical decisions. Algorithms can unlock clinically relevant information hidden in the vast volume of healthcare data guided by clinically relevant questions asked by healthcare professionals. In clinical trials of most medical studies, specific hypotheses or questions are clearly defined, and algorithms are trained to predict certain outcomes, given that a set of features and insights can be developed through prediction [41].

Artificial intelligence is used in many different health conditions and healthcare applications, for example, to assess the readmission of patients to hospitals [42], to detect patients who will suffer adverse effects when taking medications [43], in chronic diseases such as diabetes [44] and hypertension [45]; just to name a few. Table 1.1 presents some research that has applied artificial intelligence in predictive medicine.

Table 1.1 Summary of research focusing on predictive medicine
Full size table

1.4.2 Participatory Medicine

Participatory medicine focuses on the patient, in order to empower people to make more informed decisions related to their health, where the individual/patient is at the center of any related health care initiative [61]. In recent times, the role of patients in managing their own health has been changing [62]. People have gone from being mere observers to more active, committed and empowered actors in the decision-making process shared with the health team [63, 64]. For example, vaccination is strictly participatory since the individual is responsible for administering it, including that of children. This issue has become highly relevant due to the pandemic, and the multiple recently developed COVID-19 vaccines, given their differences in composition, timing of administration, effectiveness, and potential adverse events [65].

AI applied to participatory health involves individuals and their representatives, who would be the main beneficiaries of this technology [66, 67]. People can collect, record, and track indicators to quantify their health, providing insights into their overall health status [68], giving AI the opportunity to further understand their health.Continuing with the example in the pandemic context, thanks to AI, surveillance, monitoring of close contacts, compliance with quarantines, and the reduction of workload for health teams have been facilitated. Along with this, mobile technology has been harnessed to collect data on the symptoms reported by individuals combined with other data from different sources to obtain information on the disease and the spread of infection, contributing to knowledge about the prevalence and transmission of asymptomatic COVID-19 [69]. Images from security cameras, facial recognition, bank card records, and GPS data from vehicles and mobile phones have also been used to closely monitor possible contacts between infected people. In this way, data on the location and movements of people were collected to alert neighbors/patients to new cases.

AI applications within participatory health have been evaluated in the context of a wide variety of health conditions, health promotion activities, and healthcare settings through: social media, mobile apps, chatbots, health records, questionnaires, and open collaborative health studies [64, 67]. Each allows for the generation, production, or collection of various types of data, reflecting various facets of health, including psychological, physical, social, physiological, or cognitive indicators.

Tables 1.2 and 1.3 present the tools that have been used to apply AI in participatory medicine: papers that work with social networks and research where different devices and applications are used.

Table 1.2 Summary of some outstanding papers that apply artificial intelligence in social networks associated with participatory medicine
Full size table
Table 1.3 Summary of some outstanding papers that apply artificial intelligence with different devices/applications, associated with participatory medicine.
Full size table

1.4.3 Personalized Medicine

Personalized medicine is the tailoring of therapies to defined subsets of patients based on their likelihood to respond to therapy or their risk of adverse events [85]. It uses an individual’s genetic profile to guide decisions about disease prevention, diagnosis, and treatment. Knowing a patient’s genetic profile can help clinicians select the appropriate drug or treatment and administer it at the appropriate dose or regimen [86].

Personalized medicine has the potential to tailor therapy with the best response and the largest safety margin to ensure better patient care. By enabling every patient to receive early diagnosis, risk assessment and optimal treatment, it promises to improve care while lowering costs [87].

The focus of personalized medicine is on identifying which approaches will be effective for which patients based on genetic, environmental, and lifestyle factors. If information is collected from different sources for the same individual, a specific profile of the subject can be established, in which the sensitivity to certain risk factors can be accurately determined, and with this, appropriate measures can be taken to reduce the risk of developing the disease. Furthermore, risk assessment can benefit from precision medicine by generating estimates that lead to evidence-based recommendations for different outcomes [88], including historical data [89]. In addition, precision medicine’s focus on the individual will lead to accumulating evidence of multiple exposures and thus characterizing the body’s short- and long-term response to changes that occur, exposures to environmental factors, and occupational agents [88].

The use of Artificial Intelligence techniques applied to personalized medicine can contribute to the detection or prediction of diseases, achievement of accurate diagnosis, and optimal treatment. Some examples are based on personalized diagnosis and treatment based on Cardiac Imaging [90], applications of artificial intelligence in early detection of cancer [91] (such as automatic classification of cancer based on images [92] and on genes [33]), computer-aided detection and computer-aided diagnosis of colorectal polyps [93], and automatic health-record heart failure with preserved ejection fraction patient phenotyping [33]. Finally, it is highlighted that in recent times, the pandemic prompted the execution of research in personalized medicine applied in the timely diagnosis of COVID-19, in order to be able to deal with the disease, from containing the spread of the virus to the correct treatment of patients [34].

Table 1.4 presents some studies that focus on the use of artificial intelligence to support personalized medicine.

Table 1.4 Summary of research where personalized medicine is applied to deal with different problems in the field of health
Full size table

1.4.4 Preventive Medicine

Due to rising health care costs, early disease prevention has never been more important than it is today. Given the above, the concept of preventive medicine is strengthened, which is based on Medical practices that are designed to avert and avoid disease[108].

Digital health announces that thanks to technological progress embodied in applications, wearable technology, remote monitoring, telemedicine, and communication tools, among other diagnostic devices; the quality of patient care will be optimized, as well as a more timely response to any situation. This occurs because these tools make data available, which are extracted and used for the detection, prediction, and support for diagnosis/decision making of anomalies [109].

AI-driven health interventions fall into four categories relevant to global health researchers: (1) diagnosis, (2) patient morbidity or mortality risk assessment, (3) disease outbreak prediction and surveillance, and (4) health policy and planning [110]. Through the use of AI, huge amounts of data derived from different sources can be interpreted to facilitate diagnosis and increase the ability to execute early initiatives and prevent diseases to reduce the burden on both the patient and caregivers. Then executing successful interventions depends on the knowledge of the causes of the disease, the transmission dynamics, the identification of risk factors and groups, the methods of early detection and treatment, the implementation of these methods and the continuous evaluation and development of the methods. prevention methods and treatment. The identification of the determinants of a disease at the population level is a prerequisite, to prevent its appearance in a single individual, providing the basis for targeted and specific treatment [109].

The application of AI in disease prevention has been carried out in the evaluation of breast cancer risk [111], heart disease diagnosis [112], early prognosis in patients with liver disorders [113], prediction of admissions, complications and mortality in ICUs [114], and the prognosis of severity and outcome of cerebrovascular accidents [115], among others.

Table 1.5 shows a summary of some research that applies Machine Learning and Deep Learning algorithms in the field of preventive medicine.

Table 1.5 Summary of research works that seek the prevention of diseases/events in the field of health.
Full size table

1.5 Summary

People’s healthcare is changing, going from being a task focused on curing diseases to one based on preserving health, through various methods that anticipate possible illnesses, treatments to maintain indicators of internal biology of the human body within certain margins and of course, considering that people are the center of any initiative regarding healthcare. Hence, the adequate collection of quality data related to people’s lives and health is very important. For several years, enormous efforts have been made to improve the processes of capturing and storing data related to health, its subsequent maintenance, transmission, and exploitation. However, despite all these efforts, in many places there is still no unified electronic medical record, which is essential to apply the concepts behind 4P medicine. But it is a matter of time before the situation gradually improves. For now, efforts have been focused on using the data related to people’s health in the best possible way to create various applications and systems that are aimed at gradually improving population’s health. Considering this scenario and the inescapable fact that health-related data only grows in complexity and size, it is essential to have automated tools for its processing to extract information and knowledge to support decision-making in the field of healthcare. This is where artificial intelligence plays an essential role, enabling massive data processing, process automation, timely delivery of information and knowledge and, of course, support for diagnosis. We are at the gates of a new medicine, much more participatory, predictive, preventive and personalized, where the joint work between health personnel and artificial intelligence is bearing increasingly significant fruits. But remember, with all this tremendous technological power comes great responsibility. The processing of data related to people’s health must be carried out in a safe, responsible and, above all, ethical manner, taking care to safeguard the human rights of individuals above all else. Only in this way can the benefits of AI be accessed in the field of healthcare.