Магистерский диплом (ВКР) на тему Artificial Intelligence in Healthcare: opportunities and challenges in medical practice. (Работа нужна на английском)Искусственный интеллект в здравоохранении: возможности и вызовы в медицинской практике

Содержание:

ABSTRACT 2

INTRODUCTION 3

Motivation for thesis 3

Research objectives and research questions 4

Expected contribution of the research 4

CHAPTER 1. THEORETICAL BACKGROUND: ARTIFICIAL INTELLIGENCE OVERVIEW 6

1.1 Importance of Artificial Intelligence 6

1.2 Importance and Development of Artificial Intelligence Decisions 11

1.3 Risks and concerns about Artificial Intelligence 27

CHAPTER 2. ARTIFICIAL INTELLIGENCE IN MEDICINE 33

2.1 Medical issues 33

2.2 Health information management workforce 40

2.3 Transformation of healthcare professions 42

CHAPTER 3. RESEARCH METHOD 45

3.1 Method and data collection process 45

3.2 Qualified research articles 59

3.3 Data analysis 62

CHAPTER 4. RESULTS 73

4.1 Present status of Artificial Intelligence in healthcare system 73

4.2 Advantages of Artificial Intelligence in medical field 74

4.3 Scopes and challenges of Artificial Intelligence in healthcare 75

CHAPTER 5. DISCUSSION 76

5.1 Discussion of findings 76

5.2 Prospective 77

5.3 Ethical consideration 77

Conclusion 79

References 81

Appendix A. Proposed in this thesis consent form 87

Appendix B. Proposed questions for the interview 88

Appendix C. Interviewees’ information 90

Введение:

Motivation for thesis

The development of computer technology quite naturally leads to an increasing introduction of information technology in various spheres of life in modern society. An important area of research at the intersection of mathematics, informatics and biology is artificial intelligence (hereinafter referred to as AI). This is the name of a set of methods and tools designed to solve weakly formalized tasks using a computer and built according to the principles of biological systems (in particular, the human brain). Today, projects are being developed to use artificial intelligence tools to solve economic, technical, etc. problems. Of course, the field of medicine is no exception, where in many cases approaches based on existing AI technologies are used. Here you can give examples of recognizing diseases based on the results of analyzes (for example, tumors from x-rays), but AI can also help in optimizing processes related to the management of the medical industry. Thus, the introduction of artificial intelligence methods in the field of medicine is an urgent task for a specialist in the field of management, which is why it was chosen as the basic topic in this study.

Generally speaking, the subject of artificial intelligence is very extensive, but all such studies can be divided according to the type of problem being solved and the mathematical apparatus used. In particular, for better process management in the medical field, it is possible to carry out various classifications (classification is one of the typical tasks solved using AI) of patients with their further differentiated service. At the same time, the more accurately the characteristics of a particular patient are determined, the more suitable service option for him in a medical organization can be applied. Signs for classification may relate to the severity of his diseases, the financial condition of the patient, the propensity to undergo treatment and contact with doctors, etc. As is known, the problem of classification can be solved only in the case when the classes of objects are already selected in advance, which is not always fair. When it is difficult to identify stable classes in the total mass of objects (which in this work will be patients of a medical organization), the very task of identifying such classes can be addressed to artificial intelligence, and then it will be called clustering. So, the solution of these two problems (first — clustering on an array of existing patients, and then classifying new patients) can significantly improve the parameters of customer service in a medical center and improve the parameters of this medical business as a whole. Therefore, this work is devoted to the solution of such problems.

Research objectives and research questions

The aim of the work is to improve the mechanisms for dividing patients of medical centers and other medical institutions into groups (classes), which may be useful to increase the efficiency of their service in these institutions, as well as to search for potentially problematic patients or their groups (which may be important for the purposes of ensuring business stability) and can be achieved through the use of artificial intelligence technologies. To achieve the goal, it is necessary to work out the following particular research tasks:

— to analyze the concept of object clustering and consider existing methods for grouping patients of medical centers;

— based on the chosen approach from the field of artificial intelligence, develop a method for clustering patients of medical centers and institutions;

— choose one of the applied software tools for creating artificial neural networks (without developing a program in general-purpose languages, which corresponds to the work of programming specialists, but not in management, as in this study) and implement the proposed method;

— evaluate the effectiveness of the proposed method (technical, economic, social, medical, etc.).

The object of the study is the process of grouping patients of medical centers.

The subject of the study is the methods and means of clustering that can be applied and improved to group patients of medical centers.

Expected contribution of the research

The expected contribution can be assessed in several ways: the theoretical value of the work, its practical significance, as well as the potential opportunities associated with the development of this work in the future.

The novelty of the work lies in the development of a method for clustering patients of medical centers based on the use of an artificial neural network of direct propagation, which makes it possible to automate this activity and significantly reduce the labor costs required to carry out the corresponding actions by a person manually.

The practical significance of the work lies in the creation of a software tool (based on one of the existing software products for modeling artificial neural networks) with which, based on the input of a number of characteristics of a patient of a medical center, it is possible to automatically assign him to one of the classes of splitting their entire set.

In the future, this study can be developed into a separate software product that has an automated subsystem for entering information directly from the information system of the medical center with a direct connection to its database, which will make it possible to cluster patients in a fully automatic mode and quickly obtain the necessary result with a minimum of manual labor of the operator. system (in this work, data from electronic patient records were transferred manually to the developed classification system).

The methods used in the work belong to the field of cluster analysis in the part that concerns the use of artificial intelligence tools, namely the theory of artificial neural networks. The study also involved general scientific methods of analysis and synthesis.

Заключение:

Thus, this paper considers the actual problem of using the means and methods of artificial intelligence in the medical branch. This problem is important in view of the fact that, on the one hand, the capabilities of AI systems are becoming wider every year and it is completely unreasonable to ignore them in the field of medicine (while AI is being introduced more and more intensively in other areas). On the other hand, the medical industry itself contains many points where AI can and should be applied. Thus, in this paper, the concept and essence of AI technologies at the present stage of its development are considered in detail. It has been established that the most commonly used tool is an artificial neural network — a model built on the basis of reduplication in the technical system of those principles that form the basis of the functioning of the nervous system of living organisms (primarily humans).

The European Medical Center, located in Moscow, has been chosen as the base company for the research in this work. It has extremely wide capabilities for treating patients. This determines the presence of sufficiently extensive experience among the medical staff of the organization, including in the partial use of AI tools (for example, when performing operations using the Da Vinci robotic platform). Thus, this organization is a good base both for interviewing qualified personnel and for collecting depersonalized information about the center’s clients (with its subsequent analysis with AI using).

The results of the interviews showed that the staff is quite consistent in their opinion that AI in the field of operations and patient treatment can only be an auxiliary tool, while the main work should be done by an experienced human doctor. At the same time, the tasks of managing the processes occurring in the medical center can and should be managed by artificial intelligence, which allows to get more profit and improve the quality of services (convenience) for the patients of the center. Based on this opinion, it was decided to continue research on the analysis of patient data fixed in their medical records using a feed forward neural network, which will classify all the patients. The implementation of the neural network was carried out on the basis of the Matlab mathematical program, which, despite its very wide functionality, provides fairly easy-to-learn tools with graphic interface (and no coding) for modeling artificial intelligence based on neural networks (and also this software has an available full-functioning trial version for one month using).

After selecting the parameters from the patient records and generating the corresponding data set, the neural network was trained with its further validation, which showed good agreement with the results of the test sample (about 90% of matches). From a practical point of view, the network allows to determine the patient’s adherence to the use of treatment methods that require financial costs. This allows such patients, who are ready to spend significant money on their treatment, to be further offered appropriate innovative programs and approaches, while patients who like to save money will receive offers to participate in treatment activities of a more financially restrained nature (for example, including the means and methods of traditional medicine, simpler approaches of traditional medicine, etc.). Such a policy allows to increase the income of the medical center and attract patients of the first type with high-quality treatment, and also allows not to repel patients who are not ready for significant expenses with high prices at the first stage, while providing them with at least some treatment (which is clearly better than its complete absence).

The results of this research may be implemented in the specified medical EMC institution to improve the efficiency of its functioning.

Фрагмент текста работы:

CHAPTER 1. THEORETICAL BACKGROUND: ARTIFICIAL INTELLIGENCE OVERVIEW

1.1 Importance of Artificial Intelligence

There are enough high complicated problems in the modern world, which can be solved with computer using. Algorithmic complexity may varies in a very wide range: from the very simple numerous arithmetic operations (for example, a billion trivial additions per second) to extremely complex predictions methods or expert systems algorithms, etc.

If the task is simple then it is quite enough to use general mathematical methods, which we sometimes will call “classical”. Those methods were developed before the start of computer’s era, approximately before the first half of XX century, when the scientists did not know that it is possible to carry out massive calculations in a very small time period.

When the first computers appeared, researchers understood that they have very perspective instrument, which can work out much more operations per time then the most skilled human. Since then existing mathematical numerical methods evolved intensively and also new methods or even whole domains have emerged. For example in 1960s the theory of fuzzy sets (which is an important part of AI branch) was developed by Lotfi Zahde. Other extremely important AI method – using of neural networks – was invented in 1940s, but started to be developed intensively only with the development of the appropriate hardware base (from 1970s and later). As more productive computers become, as more widely neural networks are introduced for different tasks solving. So we can say that AI methods introducing is strongly associated with the development of new and more productive computers.

There is no generally accepted definition of artificial intelligence. Although there are different opinions in the scientific world about what it is, there is nevertheless no consensus. However, there is still no generally accepted interpretation of artificial intelligence, by definition, it is used to describe computer technologies, which are similar to processes related to human intelligence. The first modern name for the discipline and entity of AI was given to John McCarthy in 1956, with the other co-founders of the discipline (McCarthy et al., 1955). For its founders, it was defined as based on «the conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it» (McCarthy et al., 1955, p. 25).

Much less optimistic or straightforward, it would be defined more recently, according to the Encyclopedia Britannica, as «the ability of a digital computer or computer-controlled robot to perform tasks commonly associated with intelligent beings» (Copeland, 2019). In a similar direction, Emerj Artificial Intelligence Research, a renowned consulting firm specializing in AI, attempted to provide a lively and inclusive definition of AI by analyzing the definitional attempts of various experts and then synthesizing them to define AI as «an entity (or collective set of cooperative entities), able to receive inputs from the environment, interpret and learn from such inputs, and exhibit related and flexible behaviors and actions that help the entity achieve a particular goal or objective over a period of time» (Faggella, 2018).

Artificial intelligence is defined as “the theory and development of computer systems capable of performing creative tasks that have traditionally been performed by humans” (Haenlein M., Kaplan A., 2019). Associated with artificial intelligence is the concept of machine learning, which allows computer systems to learn from examples and build new algorithms for data processing on their own. In connection with the development of neural networks and technologies of «deep learning» (deep learning), the quality of data processing is increasing every year.

Numerous scientists have also proposed a definition of AI. Some of them are Barr and Feigenbaum, the scientists in the field of computational theory. In the early 1980s they suggested the following definition of AI: «Artificial intelligence is a field of computer science that deals with the development of intelligent computer systems, that is, systems with capabilities that we traditionally associate with the human mind,— understanding of language, learning, the ability to reason, solve problems, etc. (Barr and Feigenbaum, 2009, p. 8).

Jeff Bezos (CEO of Amazon), writes about AI as follows: «Over the past decades, computers have automated many processes that programmers could describe through precise rules and algorithms. Modern machine learning techniques allow us to do the same with tasks for which it is much more difficult to set clear rules» (Ognjanovic, 2020, p. 12).

Currently, artificial intelligence includes various software systems, together with the methods and algorithms used in them, whose main feature is the ability to solve intellectual problems in the same way as a person thinking about their solution would do (Haenlein and Kaplan, 2019). Among the most popular areas of AI application are forecasting, evaluating any digital information with an attempt to give a conclusion on it. Artificial intelligence, one of the capabilities of which is able to force a computer or a computer-controlled robot to perform tasks related to intelligent beings. It is widely used in the development of a system that has the characteristics of a human intelligence process. For example, learning from experience, reasoning, distinguishing. As the technology is developing, computer programs perform as good as specialized persons, like in computer search, handwriting recognition, medical diagnosis and so on. (Copeland 2019).

In the 21st century, AI techniques have experienced a resurgence following advances in the computing power of computers, big data and theoretical advances in the field. This is what led to the prowess presented earlier. Moreover, AI techniques have become an essential part of the healthcare sector, helping to solve many complex problems in medicine (Clark, 2015).

Through the years, much research has been done and many definitions of AI have been provided in the literature. Therefore, in table 1.1 some of the concepts are presented to make an overall vision of what Artificial Intelligence is.

Artificial Intelligence can be interpreted differently, and its influence over the time can be traced even better. The impact of AI and its importance cannot be underestimated on various aspects of our lives. The interpretation of Barr and Feigenbaum can be seen as the most revealing concept describing the actions that could be learned and implemented into practice. Their idea of reflecting a computer in relation to the human mind can be perceived as the most applicable to a person, which means that his further interaction will take place with people in practice and therefore will be considered the most tied to the human factor.

AI and mentioned methods of its using (and much more others) allow end-users to get answers to various practical problems. In general AI methods are designed to solve weakly formalized (or non-formalized) tasks. Formalized tasks are relatively simply can be solved with using some mathematical algorithms but it is very hard (or even impossible) to develop an algorithm for solving weakly formalized task. For example, if researcher knows that some set of parameters are relevant for some output value, he must build some formula (mathematical model) which connects those parameters with the output value. This task is complex enough and sometimes it is not possible to get an appropriate decision. AI methods usually do not need building of the mathematical model, but only of using large sets of data: concrete values of parameters and corresponding output values. AI systems are trained with using those data-sets and do not require from the researcher to develop some models. The only requirement is to define an appropriate structure and some internal rules for the AI system but the other work it will work out by itself.

Often AI methods replicate or at least are based on some physiological mechanisms in a human (or some highly developed animals) body. So in such a context AI is related to bionics scientific branch which

With using of the AI approaches can be solved next tasks:

— classification or pattern recognition in a wide sense (finding of the most appropriate class for some object from some definite set of classes);

— classification or recognition of images (graphical pictures);

— clustering / categorization (i.e., dividing the set of elementary objects into separate clusters or categories; in other words: finding a set of classes which are used for the classification);

— mathematical approximation of functions (can be used as the basis for other practical methods);

— prediction / forecasting;

— optimization;

— organization of associative memory;

— control of carious technical and technological processes (provided by the corresponding machines with CNC).

Let’s consider these tasks in more detail.

In pattern classification or recognition, the task is to establish whether an input signal or “image” (for example, a speech signal or a handwritten symbol), represented by a certain features vector, belongs to one or more predefined classes. This includes the most famous, one might say classic, problem solved with the help of neural network technology — recognition of individual letters, to which very many researches are dedicated. This class of tasks also includes speech recognition, electrocardiogram signal classification, blood cell classification (both are related to medicine domain) and so on.