Computational Retinal Image Analysis

Tools, Applications and Perspectives

Emanuele Trucco, Tom MacGillivray, Yanwu Xu

• Publisher: Academic Press
• Year: 2019
• ISBN:
  9780081028179
• Topic: Computer Science
• Subtopic: Digital Media
• Index: Computer Science

Computational Retinal Image Analysis: Tools, Applications and Perspectives gives an overview of contemporary retinal image analysis (RIA) in the context of healthcare informatics and artificial intelligence. Specifically, it provides a history of the field, the clinical motivation for RIA, technical foundations (image acquisition modalities, instruments), computational techniques for essential operations, lesion detection (e.g. optic disc in glaucoma, microaneurysms in diabetes) and validation, as well as insights into current investigations drawing from artificial intelligence and big data. This comprehensive reference is ideal for researchers and graduate students in retinal image analysis, computational ophthalmology, artificial intelligence, biomedical engineering, health informatics, and more.

• Provides a unique, well-structured and integrated overview of retinal image analysis
• Gives insights into future areas, such as large-scale screening programs, precision medicine, and computer-assisted eye care
• Includes plans and aspirations of companies and professional bodies

Chapter 1

A brief introduction and a glimpse into the past

Emanuele Trucco^a; Yanwu Xu^b; Tom MacGillivray^{c a} VAMPIRE Project, Computing (SSEN), University of Dundee, Dundee, United Kingdom
^b AI Innovation Business Department, Baidu Online Network Technology (Beijing) Co., Ltd., Beijing, China
^c VAMPIRE Project, Centre for Clinical Brain Sciences, University of Edinburgh, Edinburgh, United Kingdom

Abstract

This chapter gives a brief introduction to the book, its purpose, content and structure. It also contains a concise glimpse into the origins of eye- and vision-related research, which date back at least to the time of ancient Greece.

Keywords

Retinal image analysis; History of eye research; History of vision research

Acknowledgments

Our sincere thanks go to the Elsevier team for their assistance with the production of this book from an idea to publication, especially Tim Pitts and Mariana Kuhl, including for unshakeable patience with our multiple infringements of deadlines.

We thank the many colleagues who have authored the chapters forming this book. We hope that they, and indeed all readers, will find it useful and worth recommending for many years to come.

Finally, thank you to you, the reader, for buying this book. We hope you will find it a valuable resource for teaching and research alike. We welcome your feedback.

1 Why this book?

This book offers an overview of the main aspects of contemporary retinal image analysis (RIA) in the context of clinical applications, healthcare informatics and artificial intelligence. The book aims to be not just another collection of papers on technical advancements; these are reported more timely by conferences papers, online journal pre-prints and repositories like arXiv or bioRxiv. Instead, the book aspires to be a comprehensive introduction to the field. A logical progression of chapters takes the reader through an overview of RIA, its clinical motivations, technical foundations (image acquisition modalities, instruments), computational techniques for essential operations (e.g., anatomical landmarks location, blood vessel segmentation), lesion detection (e.g., optic disc in glaucoma, microaneurysms in diabetes) and the important topic of validation, all the way to a showcase of current investigations drawing from artificial intelligence and big data (retinal biomarkers for risk of systemic conditions) and the future, e.g., large-scale screening programs, precision medicine, computer-assisted personalized eye care and the challenges of creating, maintaining and making available to research inceasingly large collections of clinical data.

2 Casting an eye into the distant past: The history of eye research in the West

Zusammengestohlen aus verschiedenem diesem und jenem^a

Ludwig van Beethoven, comment on his String Quartet no. 14

For an overview of retinal imaging including a brief history of modern techniques we refer the reader to Keane and Sadda’s review [1]. Chapter 3 of this book offers a contemporary view on retinal imaging. We add here a brief summary of the origin of eye-related research in the West, which we hope may interest the reader; the present cannot be understood completely but in the context of the past. Beyond the western world, a concise history of ancient vision and eye-related theory in India and some comparisons with western theories is given by Deshpande [2]. Our short account follows loosely Pierantoni’s detailed book [3], with elements from the history of ophthalmology (anatomy and physiology) available on the UK Royal College of Ophthalmologists website [4]. Bynum [5] is a concise, non-technical but informative account of the history of medicine, including of course ophthalmology.

The curiosity of man for the eye goes back a long way. It may seem therefore extraordinary that the first anatomically accurate drawing of the ocular bulb did not appear before the early 17th century, in Rosa Ursina by the German Jesuit priest and scientist Christoph Scheiner (1575–1650). Before then, all drawings of the eye put religious or philosophical beliefs before anatomical observation. This fascinating story starts, in the West, with the oldest known drawing of the ocular bulb and its main component, due to the medieval arab scholar Hunain ibn Ishak, who lived in the 9th century BC, in his “Ten essays on the structure of the eye, its diseases and cures.” The drawing was first reproduced by the neurologist Stephen Polyak during WWII [6] and is itself a copy of an older drawing which did not reach us, perhaps a Greek manuscript from many centuries before. Ibn Ishak’s eye looks very inaccurate to us: it is almond-shaped; the lens is in the center; the optic nerve, the pupil and the lens are aligned along the middle axis of the ocular bulb; the optic nerve is hollow. Vision is explained through the flow of a “vital spirit” emitted by the eye (an idea already found in Pythagoras and Euclid), which required the hollow optical nerve to flow out of the eye and back. As Pierantoni observes, the drawing is best interpreted as a functional diagram, as all functional elements are in place, but not anatomically accurate. Yet dissecting an eye does not require particularly sophisticated instruments. But the influence of the philosophical giants of the antiquity, in addition to the aversion to cadaver dissection of religion and state, was obviously immense—and would last for many centuries to come.^b Hence dissection was often practiced on animals, assuming that their anatomy was similar or the same to the human one.

The first drawing showing a connection between the eye and the brain is due to another arab scholar, Abu Ja’far Ahmad ibn Muhammad (906–963), again most likely a copy of a previous Greek document. This drawing is also the first diagram giving an account of binocular vision: it shows two ocular bulbs, now round but still equipped with hollow channels carrying the “visual spirit” responsible for vision. A relation of Ja’far, the scholar Ibn al-Haitam (also known as Alheizen, c.965–1040), proposed however in his Book of Optics that vision was made possible by the rays of light entering the eye. Precursors of this idea were already present in the work of the highly regarded Greek physician and surgeon, Galen (129–c.200 AD), who thought that the light entering the eye interacted with a “visual spirit” (pneuma) generated in the brain. The pneuma would carry the shapes carried by the light to the brain, in a bi-directional flow. In contrast, early functional theories of vision, including those by Pythagoras and Euclid, who lived between the mid-5th and the mid-3rd century BC, stated that the pneuma was emitted into the world and would bounce off objects carrying back their shapes into the eye. Democritus (4th–3rd century BC) had raised a dissenting voice, postulating that objects emitted continuously images of themselves (éidola, or “figures,” “representations”), that entered the eye (intromission theory) making perception possible.

The eye model undergoes an idealization in the Middle Ages, seemingly to reflect more the geometry of the divinely perfect, symmetric universe, in which the circle represented the supreme geometric perfection. The anatomy of the eye was therefore inspired by the orbits of the known planets, that had to be rigorously circular. Such models are found in the work by Roger Bacon (1214–92) and John Pecham (1230–92), Archbishop of Canterbury. An eye model based on tradition, not anatomy, is found even in Leonardo da Vinci’s well-known section of a man’s head in his Anatomical Studies, folio 32r. Here the eye is depicted as a spherical bulb containing a central, spherical lens. Leonard’s eye is connected to the brain by a channel that he believed was composed by many smaller ones, to keep separate the images of the different things perceived simultaneously. Leonardo never took a definite position in the controversy between emission and intromission theories.

Dissection and direct observations re-start in earnest with Andreas Vesalius, the Flemish physician regarded as the father of modern anatomy. Vesalius’s drawing of the eye in his De humani corporis fabrica (on the structure of the human body, 1543) suggests direct observation (e.g., the bulb is spherical, the anterior chamber is present) but still reflects the burden of the tradition: the lens is central, the optic nerve is hollow and aligned with the central axis of the bulb through the center of the pupil.

In the work by Johann Kepler (1571–1630), the German astronomer and physicist, optics plays a crucial role and allows Kepler to propose an explanation for the paradox of the inverted image, which had puzzled Leonardo himself: if the pupil, as a small aperture, makes the image appear inverted on a screen, why don’t we perceive the world upside down? Kepler understands that an “opposite inversion” must happen in the brain. Finally, shortly after Kepler, Scheiner publishes the first anatomical drawing of the eye which we can accept completely. A correct understanding of the anatomy was finally achieved. Centuries would still be needed to attain accurate physiological models, but this goes beyond the scope of this short historical note.

3 Book structure

The book is organized in five parts following logically from each other a...

Table of contents

• Cover image
• Title page
• Table of Contents
• Copyright
• Contributors
• Chapter 1: A brief introduction and a glimpse into the past
• Chapter 2: Clinical motivation and the needs for RIA in healthcare
• Chapter 3: The physics, instruments and modalities of retinal imaging
• Chapter 4: Retinal image preprocessing, enhancement, and registration
• Chapter 5: Automatic landmark detection in fundus photography
• Chapter 6: Retinal vascular analysis: Segmentation, tracing, and beyond
• Chapter 7: OCT layer segmentation
• Chapter 8: Image quality assessment
• Chapter 9: Validation
• Chapter 10: Statistical analysis and design in ophthalmology: Toward optimizing your data
• Chapter 11: Structure-preserving guided retinal image filtering for optic disc analysis
• Chapter 12: Diabetic retinopathy and maculopathy lesions
• Chapter 13: Drusen and macular degeneration
• Chapter 14: OCT fluid detection and quantification
• Chapter 15: Retinal biomarkers and cardiovascular disease: A clinical perspective
• Chapter 16: Vascular biomarkers for diabetes and diabetic retinopathy screening
• Chapter 17: Image analysis tools for assessment of atrophic macular diseases
• Chapter 18: Artificial intelligence and deep learning in retinal image analysis
• Chapter 19: AI and retinal image analysis at Baidu
• Chapter 20: The challenges of assembling, maintaining and making available large data sets of clinical data for research
• Chapter 21: Technical and clinical challenges of A.I. in retinal image analysis
• Index