Artificial Intelligence and Ophthalmology: Perks, Perils and Pitfalls

The book helps to explore the vast expanse of artificial intelligence-based scientific content that has been published in the last few years. Ophthalmology has recently undergone a silent digital revolution, with machine learning and deep learning algorithms consistently outperforming human graders in studies published across the globe. It is high time that a resource that breaks this information behemoth into easily digestible bits comes to the fore. This book simplifies the complex mechanics of algorithms used in ophthalmology and vision science applications. It also tries to address potential ethical issues with machines entering our clinics and patients’ lives. Overall it is essential reading for ophthalmologists/eye care professionals interested in artificial intelligence and everyone who is looking for a deep dive into the exciting world of digital medicine. 

• Language: English
• Publisher: Springer
• Release date: Apr 22, 2021
• ISBN: 9789811606342

Book preview

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021

P. Ichhpujani, S. Thakur (eds.)Artificial Intelligence and OphthalmologyCurrent Practices in Ophthalmologyhttps://doi.org/10.1007/978-981-16-0634-2_2

2. What You Need to Know About Artificial Intelligence: Technical Introduction

Oscar J. Perdomo¹  , Santiago Toledo², Alvaro Orjuela¹ and Fabio A. González²

(1)

Universidad del Rosario, Bogotá, Colombia

(2)

Universidad Nacional de Colombia, MindLab Research Group, Bogotá, Colombia

Oscar J. Perdomo

Email: [e-mail]

Keywords

Artificial neural networksEvolutionary algorithmsFuzzy logicSupport vector machinesArtificial neural networks

2.1 Artificial Intelligence

The popularity of artificial intelligence (AI) has increased in different fields of application. Initial terminology pertaining to AI was proposed in the middle of the last century. However, in the nineties decade, expressions like expert systems, artificial neural networks, fuzzy systems, and others were based on statistical tools from the 60 and 70 decades, creating the field known as machine learning (ML). Currently, these topics are associated with automation applications and an emergent area of data science. Different processes have applied these technological tools thereby improving the solutions to problems in distinct fields and at different levels. In spite of the increment in the use of the terms AI and ML, it is difficult for many to determine what exactly AI is.

The Institute of Electrical and Electronics Engineers (IEEE) subdivides the AI field into three subfields:

1.

artificial neural networks (ANN), which are based on connectionist models, trying to emulate the biological brain;

2.

evolutionary algorithms that employ bioinspired methods of optimization as, for example, the mechanism of natural selection; and,

3.

fuzzy logic, which use the natural language in human being, modifying the classical logic.

These paradigms have been included inside the concept of computational intelligence (CI). However, the similarities between AI and CI are notable, CI emerged from a community virtually different [1]. Since there is no unification, the terminology is wide and, currently, employ concepts related to the explanation as to how the machines learn [2]. Therefore, the main aspect of CI is the numeric representation of the knowledge compared to the symbolic representation of the AI.

Simultaneous to the development of the CI community, on the other side, Vapnik and Chervonenkis proposed more models that learn from data, which were coined as machine learning [3]. Mainly they used different statistical tools for different strategies to solve problems in classification and regression. The popularization of these models gave birth to the new field of ML.

The ML area is composed of models such as support vector machines (SVM), trees for regression and classification (RT), and ANNs, just to mention a few. More recently, Rosenblatt’s work based on perceptron evinced improvements in the area, and later with the multilayer perceptron (MLP) and the backpropagation (backprop, BP) algorithm from McLellan [4] has exhibited its splendor.

Neural networks were reborn with the addition of layers to the ANN models. This field has been named deep learning (DL), and currently, it is one of the most popular methods to solve challenges in image processing and computer vision [5]. The number of parameters of the neural network has been increased with additional problems in the training model, where more synaptic weights have to be tuned, demanding more computation capacity and time processing. Figure 2.1 shows the association of the models in terms of AI as a big technological and study area and ML as a subfield of the AI and the DL as a particular scenario of the