Artificial Intelligence in Precision Health

From Concept to Applications

Debmalya Barh

• Publisher: Academic Press
• Year: 2020
• Print ISBN:
  9780128171332
• eBook ISBN:
  9780128173381
• Topic: Medicine
• Subtopic: Pharmaceutical, Biotechnology & Healthcare Industry

Artificial Intelligence in Precision Health: From Concept to Applications provides a readily available resource to understand artificial intelligence and its real time applications in precision medicine in practice. Written by experts from different countries and with diverse background, the content encompasses accessible knowledge easily understandable for non-specialists in computer sciences. The book discusses topics such as cognitive computing and emotional intelligence, big data analysis, clinical decision support systems, deep learning, personal omics, digital health, predictive models, prediction of epidemics, drug discovery, precision nutrition and fitness. Additionally, there is a section dedicated to discuss and analyze AI products related to precision healthcare already available. This book is a valuable source for clinicians, healthcare workers, and researchers from diverse areas of biomedical field who may or may not have computational background and want to learn more about the innovative field of artificial intelligence for precision health. - Provides computational approaches used in artificial intelligence easily understandable for non-computer specialists - Gives know-how and real successful cases of artificial intelligence approaches in predictive models, modeling disease physiology, and public health surveillance - Discusses the applicability of AI on multiple areas, such as drug discovery, clinical trials, radiology, surgery, patient care and clinical decision support

Section II

Applications of artificial intelligence in precision health

Chapter 7

Predictive models in precision medicine

Göksu Bozdereli Berikol^a,b; Gürkan Berikol^{a,b a} Department of Emergency Medicine, Istanbul Bakirkoy Dr. Sadi Konuk Training and Research Hospital, Istanbul, Turkey
^b Department of Neurosurgery, Karaman Public Hospital, Karaman, Turkey

Abstract

The curiosity for why and how events occur, and what might have caused them necessitated calculation of prediction. Predictive models are the milestones of artificial intelligence. Predictive analysis not only encompasses predictive modeling, but also some other fields like data mining and machine learning. Predictive analysis is composed of the steps: data collection, data analysis, and statistical analysis, predictive modeling, and imaging outcomes. In this chapter, we aimed to define the predictive models and analysis with the advantages and disadvantages, and illustrate how it works for precision medicine on a model-based approach.

Keywords

Predictive modeling; Predictive analysis; Precision medicine; Machine learning; Expert systems; Neural networks

Introduction

The curiosity for why and how events occur, and what might have caused them necessitated the calculation of prediction. Predictive models are the milestones of artificial intelligence. However, physicians’ experiences, advancing technologies and changing literature guide diagnosis and treatment of diseases, the constellation of them, when combined with prediction analyses, creates artificial intelligence and decision support systems assisting physicians. Predictive analysis not only encompasses prediction modeling, but also some other fields like data mining and machine learning. Predictive analysis is composed of the steps of data collection, data analysis, and statistical analysis, predictive modeling, and imaging outcomes. The most important step among them is data collection and processing. The predictive calculations are based on probability. Calculation of multiple probabilities is based upon clustering or classification of those probabilities. As Hippocrates once said, “It is important to know what sort of person has a disease than to know sort of disease a person has,” the results of prediction are personal. Although disease diagnosis is global, a person’s genetic and epigenetic factors affect the outcome. The role of personalized medicine is to predict a person’s tendency to a particular disease, or the natural course or treatment response of that disease, using the perfect molecular structure of each person. Physicians reinforce the knowledge they acquire in medical faculty with that they acquire through their personal experiences in their professional life. It maintains its popularity in all medical fields, mainly oncology, and predictive modeling aids precision medicine in all these fields.

Predictive analysis

Estimation is predicting unknown outcomes. Predictive analysis, on the other hand, encompasses predictive modeling, but also other fields like data mining and machine learning. Predictive analysis serves to obtain information about likely outcomes of target events (Nyce and Cpcu, 2007; Eckerson, 2007). The difference in predictive analysis from probability calculations is that it also incorporates unpredictable processes and behaviors into the process. In addition to statistical methods, it also makes use of machine learning experienced through training data.

Machine learning has two types: unsupervised and supervised learning. Complexed patient data can be used with supervised and unsupervised learning. Supervised learning is configured to uncover the association between cause and effect (Şahin et al., 2018; Deo, 2015; Senders et al., 2017, 2018). Unsupervised learning applications are aimed to find out the complexed associations among this big data (Senders et al., 2017). It can be referred to make a prognosis, diagnosis with unsupervised learning (Senders et al., 2017).

Future predictions are made using certain variables such as past events, factors related to those events, how those factors affected those events, and the grade of interfactorial relations. Predictive analysis encompasses predictive models, descriptive models, and decision models.

Predictive analysis consists of data collection, data analysis, statistical analysis, predictive modeling, and visualization of outcomes. Among them, data collection and processing are the most important ones. There are some examples like magnetic resonance imaging of tumor size changes between preoperative and postoperative, and prediction of mortality from intensive care unit monitorization changes (Senders et al., 2017).

Predictive modeling

Predictive modeling is a modeling type based upon a prediction of outcomes using statistics (Geisser, 2017). The curiosity for why and how events occur, and what might cause them, created the necessity of calculation of prediction. Predictive models are used in many fields from emails to spam prediction, banking, marketing, ensuring, social media, weather forecast, criminal, and medical. Prediction calculation depends on probability. Calculation of multiple probabilities is based upon the principle of clustering or categorization of those probabilities.

Predictive models

Predictive models use multiple methods. Among these, the most commonly used ones are generalized linear models, decision trees, neural networks, support vector machines, clustering algorithms, naïve Bayes, K-nearest neighbor, random forest, logistic regression, time series analysis, and deep neural network (Finlay, 2014).

Table 1 shows the superiorities and disadvantages of the models (Suganya et al., 2015; Chen and Romanowski, 2013; Wang, 2017).

Table 1

Advantages and disadvantages of the models

	Superiorities	Disadvantages
Generalized linear models	Easy	Not suitable for complex and big data
Decision trees	Easy Performance is not affected by nonlinearity Missing values problem can be solved	Complex Subtrees can be duplicated Trees can vary due to the complexity
Artificial neural networks	Efficient in noisy data High computational rate	Poorly semantic slow The network architecture is complex
Support vector machines	Highly accurate classifies Suitable for noisy data Overfitting is solved	Slow Not suitable for multiclass classifications
Clustering algorithms	Easy	Worse accurate outcomes Now knowing the optimal numbers of clusters
Naïve Bayes	Fast Better performance	Infeasible information Infeasible computation
K-nearest neighbor	Easy and fast technique Applied to noisy data Suitable to multimodal classes	Sensitivity to the structure of the data Low memory Slowing down at supervised learning
Random forest	Better accuracy than decision trees Efficient for big data Suitable for linear and nonlinear data	Overfitting Slow Difficult interpretation for complex trees
Logistic regressions	Easy Adaptation to a new data input	Sensitive to missing and extreme values
Time series analysis	Suitable for multivariate analysis	Complex Difficulty in specifying the relations
Deep neural network	Useful in big and complex data Better performance and accuracy Fast	Overfitting Intensive in computational work
Fuzzy logic	Suitable for uncertain problems and stochastic relationships	Expert knowledge is required Results depend on the rules or decisions
Genetic algorithms	Better performance	Slow

Precision medicine

Personalized medicine deals with the prediction of a person’s disease tendency, disease course, or treatment response using the information containing a perfect molecular structure of that person (Davis et al., 2009; Barh et al., 2013; Nicholson et al., 2011).

In recent years, gov...

Table of contents

• Cover image
• Title page
• Table of Contents
• Copyright
• Dedication
• Contributors
• Editor's biography
• Preface
• Section I: Artificial intelligence technologies
• Section II: Applications of artificial intelligence in precision health
• Section III: Precision systems in practice
• Index