H Kamdar, Jeba Praba and John Georrge
Since the late 1980s, artificial intelligence (AI) has rapidly evolved, particularly in healthcare, where the surge in datasets has led to significant advancements in AI publications over the past two decades The rise in computational power has facilitated the availability of AI technologies, primarily through machine learning, which analyzes structured data such as images and genetic information, and natural language processing, which focuses on unstructured data Both methods have seen substantial improvements in their algorithms and applications, enabling the analysis of various diseases, including cancer, neurological disorders, cardiovascular issues, liver disease, and congenital cataracts Recently, the introduction of deep learning has sparked a revolution in AI, promising transformative enhancements in diagnostic medical imaging systems, including endoscopic diagnosis, pathology, and dermatology This article provides essential technical insights into the popular methods, algorithms, and applications in medical diagnosis that have emerged in recent years.
Keywords Medical diagnosisãDeep learningãMachine learningã Genetic algorithmã Cancer ãNervous system diseaseãCardiovascular diseaseã Liver disease
J H Kamdar and J Jeba Praba are considered as Joint first.
ICAR-Directorate of Groundnut Research, Junagadh, Gujarat, India
Department of Computer Applications, Christ College, Rajkot, Gujarat, India
Department of Bioinformatics, Christ College, Rajkot, Gujarat, India e-mail: johnjgeorrge@gmail.com © Springer Nature Switzerland AG 2020
V Jain and J M Chatterjee (eds.), Machine Learning with Health
Care Perspective, Learning and Analytics in Intelligent Systems 13, https://doi.org/10.1007/978-3-030-40850-3_2
Artificial intelligence (AI) is a well-defined field of science and engineering focused on computational understanding and the creation of artifacts that exhibit intelligent behavior Over the past four decades, AI techniques have been utilized in medical applications and diagnostics to enhance healthcare by improving physicians' effectiveness, efficiency, and coherence The rapid advancements in AI accuracy and the availability of large datasets have made it a preferred choice for addressing medical diagnostic challenges AI software has significantly contributed to its implementation in healthcare The applications of AI in clinical diagnosis are diverse, marking the fourth decade of its integration into medicine Key advantages of AI include accuracy, cost efficiency, and replicability The primary AI techniques are Machine Learning (ML) and Natural Language Processing (NLP), with ML analyzing structured data such as genetic and imaging data, while NLP focuses on unstructured data like clinical notes and medical journals Additionally, machine learning techniques are categorized into classical and deep learning methods This chapter provides an overview of the AI devices, algorithms, and applications that have emerged in medical diagnosis in recent years.
Machine Learning (ML) is a significant and rapidly evolving branch of Artificial Intelligence (AI) It emulates various aspects of human cognitive abilities, including planning, learning, logic, comprehension, perception, and interaction By utilizing a range of tools, AI seeks to replicate human intelligence in these domains, enabling advanced data manipulation and understanding.
Machine learning (ML) techniques utilize empirical data, such as age, gender, physical examinations, imaging, gene expression, and electrophysiology (EP), to identify underlying data structures To effectively analyze these datasets, three primary algorithms are employed: the Supervised Algorithm, which guides learning with labeled data; the Unsupervised Algorithm, which identifies patterns without labels; and the Semi-supervised Algorithm, which combines both approaches to enhance learning outcomes.
Supervised machine learning (ML) relies on known classified data sets to predict relationships between specific traits and outcomes, making it particularly valuable in medical diagnosis due to its ability to yield relevant results Among the most commonly used supervised algorithms are Support Vector Machine (SVM) and Neural Networks, which utilize advanced learning techniques SVM, one of the simplest ML algorithms, is designed for binary classification tasks but can also handle multi-class data sets It works by drawing a line between two clusters to make predictions and has found significant application in the field of medical diagnosis.
Fig 1 Support vector machine (SVM) example
Neural networks are a powerful alternative machine learning method, particularly effective for non-linear classification problems They consist of three main layers: an input layer, one or more hidden layers, and an output layer Each node in the network processes inputs to generate outputs that are passed to subsequent nodes, ultimately leading to the classification of traits based on the initial input values Research by Ramesh et al highlights that neural networks are among the most commonly used diagnostic tools, with fuzzy logic–neural networks emerging as a prevalent AI technique in various fields, including genetics, cardiology, and radiology.
Deep learning, an advanced evolution of traditional neural networks, has experienced significant growth in recent years, driven by enhanced computational power and the availability of new data sets This technology enables the analysis of more complex, non-linear data, making it a powerful tool for various applications.
Fig 2 Typical neural network configuration
In recent years, deep learning has gained immense popularity due to the increasing volume and complexity of data Unlike traditional neural networks, deep learning architectures utilize multiple hidden layers, allowing them to process and learn from diverse and intricate data structures effectively.
In medical applications, various types of neural networks play crucial roles, including recurrent neural networks, deep neural networks, convolutional neural networks (CNN), and deep belief networks, which are commonly used methodologies for developing deep learning algorithms.
Recently, Convolutional Neural Networks (CNNs) have been effectively utilized in the medical field for disease diagnosis, leveraging popular software frameworks such as Microsoft’s CNTK, Google’s TensorFlow, and Berkeley AI Research’s Caffe In certain instances, the classification of data may be unfamiliar, leading to the use of unsupervised learning techniques.
Machine learning techniques play a crucial role in data categorization, with two widely used unsupervised methods for identifying significant features being K-means clustering and Principal Component Analysis (PCA).
K-means clustering is a technique used to identify groups where data points are closest to each other, while Principal Component Analysis (PCA) organizes data into a centered matrix, where each row represents a data model and each column represents a feature's value In this centered matrix, the mean of each column is subtracted from its values to ensure that the mean becomes zero Additionally, semi-supervised learning combines elements of both unsupervised and supervised learning, making it suitable for scenarios where data is incomplete for certain topics.
The Natural language processing (NLP) is a platform of computational philology and
AI promotiveness involves the analytical development of realistic language, particularly within the medical field The vast literature in this domain, including numerous studies utilizing Natural Language Processing (NLP) methods, highlights the significance of Electronic Health Records (EHR) systems, which have provided a wealth of information over the years These systems encompass electronic medical notes that enhance healthcare quality through effective disease monitoring, evidence-based medicine, and decision support To achieve reliable and efficient clinical text processing, it is crucial to implement robust NLP strategies Over the past decade, there has been a notable increase in NLP-based medical research, with 1,405 publications reflecting an average annual growth rate of 18.39%.
For individuals looking to explore AI applications across various domains, numerous online resources are available These include software that allows researchers to conduct AI experiments with their own data, statistical learning and logic inference tools, open-source data mining systems, programming frameworks for AI methodologies, and companies offering free AI services A selection of these essential tools is detailed in Table 1.
5 Application of AI in Medical Diagnosis
Over the past decade, the availability of articles, books, and datasets has significantly increased, showcasing a wide range of applications for artificial intelligence (AI) Various AI algorithms, both independently and in combination with other methods, have been thoroughly reviewed and successfully implemented in medical diagnostics, including conditions such as stroke, Alzheimer’s disease, skin cancer, and other neurological disorders.