Computational Retinal Image Analysis: Tools, Applications and Perspectives

Front matter
Copyright
Contributors
A brief introduction and a glimpse into the past
	Why this book?
	Casting an eye into the distant past: The history of eye research in the West
	Book structure
	Acknowledgments
	References
Clinical motivation and the needs for RIA in healthcare
	Introduction
		Assisting diagnosis of clinical eye diseases
		Assessing severity and classifying clinical eye diseases
		Capturing pre-clinical signs of the eye diseases
		Identifying retinal changes associated with systemic diseases
		Structural signs to functional signs
	Perspectives—Precise diagnosis, replacing repetitive work, and exploring novel signs
	References
The physics, instruments and modalities of retinal imaging
	Introduction
	Optics of the eye
		Using the eye to record images of the retina
		Spatial resolution of retinal images
		Glare, contrast and image quality
		How the physics of light propagation affects retinal image quality
		Spectral characteristics of the eye
		The use of eye phantoms to simulate retinal imaging
	Ophthalmic instruments
		Brief history
		Safety exposure limits
		The fundus camera
		Indirect ophthalmoscopes
		The scanning laser ophthalmoscopes
		Handheld retinal cameras
		Ultrawide field imaging
		Optical coherence tomography
			Time domain optical coherence tomography. The beauty of the en-face view
			Spectral domain optical coherence tomography
			Camera based optical coherence tomography and exceptional spatial resolutions
			Swept source optical coherence tomography. Going faster and deeper into the tissue
			Methods of generating images in SD-OCT
			Modern topics in optical coherence tomography for eye imaging
	Polarization and birefringence
	Conclusions
	References
Retinal image preprocessing, enhancement, and registration
	Introduction
	Intensity normalization
		Fundus imaging
		Tomographic imaging
	Noise reduction and contrast enhancement
		Fundus imaging
		Tomographic imaging
	Retinal image registration
		Fundus imaging
		Tomographic imaging
		Intramodal vs. cross-modal image registration
	Conclusions
	Acknowledgment
	References
Automatic landmark detection in fundus photography
	Background
		Optic disc
		Macula lutea
	Fovea and disc detection/segmentation—Utility
	Retinal imaging databases
	Algorithm accuracy
	Optic disc and fovea detection
		Automated localization of the optic disc, fovea, and retinal blood vessels from digital color fundus images (Sinthana ...
		Locating the optic nerve in a retinal image using the fuzzy convergence of the blood vessels (Hoover and Goldbaum, 2 ...
		Detection of optic disc in retinal images by means of a geometrical model of vessel structure (Foracchia et al., 2004 ...
		Fast localization and segmentation of the optic disc in retinal images using directional matched filtering and level ...
		Multiscale sequential convolutional neural networks for simultaneous detection of the fovea and optic disc (Al-Bander ...
	Summary
	References
Retinal vascular analysis: Segmentation, tracing, and beyond
	Introduction
	Benchmark datasets and evaluation metrics
		Datasets
		Evaluation metrics
	Vessel segmentation
		Unsupervised segmentation
		Supervised segmentation
		Deep learning
	Vessel tracing
		Vascular junction identification
		Vascular tree separation
		Arterial/venous vessel classification
		Clinical relevant vessel readouts
	Summary and outlook
		Vasculature analysis in emerging imaging techniques
		Benchmarks and metrics
	References
OCT layer segmentation
	Anatomical description and clinical relevance
	Algorithmic evaluation and benchmarking
	Intensity based methods
	Graph based methods
	Deep learning based methods
		Preprocessing and augmentation
		Pixelwise semantic segmentation methods
		Boundary detection methods
	Discussion and conclusion
	References
Image quality assessment
	Introduction
		Image quality of ophthalmic images
		Applications of image quality assessment algorithms
			Screening for diabetic retinopathy
			Teleophthalmology and clinical decision making
			Epidemiology study requirements
	Automated image quality assessment algorithms
		An overview of techniques
		Datasets and metrics used to evaluate image quality
		Examples of retinal image quality assessment systems
			Algorithms based on generic image quality parameters
				Information fusion
			Algorithms based on structural image quality parameters
				Image structure clustering
				Segmentation map feature analysis
		Algorithms based on deep learning
			Convolutional neural networks
			Human visual system information combined with convolutional neural networks
	Conclusion
	References
Validation
	Introduction: Why is validation difficult?
	Challenges
		Annotations are expensive
		Annotation tasks are often unfamiliar to clinicians
		Consistency is hard to achieve
		Collecting annotations may be limited by data governance
		Image quality may vary across images and data sets
		Absence of unambiguous ground truth
		Time-varying quantities are not well represented by a single measurement
		Test criteria and data sets are not uniform in the literature
		Dependency on application/task
		Human in the loop
	Tools and techniques
		Choosing images: Aligning data set with clinical criteria
			Technical criteria
			Clinical criteria
		Direct techniques: Focus on the image processing task
			Receiver operating characteristic (ROC) curves
			Accuracy and related measures
			Confusion matrices
			Bland-Altman graphs
			Cohen’s kappa and related measures
			Error histograms
				Eliminating outliers
				Choosing an appropriate number of bins
		Validation on outcome: Focus on the clinical task
	Annotations and data, annotations as data
		Annotation protocols and their importance
		Reducing the need for manual annotations
	Conclusion
	Acknowledgments
	References
Statistical analysis and design in ophthalmology: Toward optimizing your data
	Introduction
		Data analysis in ophthalmic and vision research
		The contribution of statistics in ophthalmic and vision research
	Data classification, data capture and data management
		Data classification
		Data collection and management
		Words of caution about data collection in the current era of big data
	Uncertainty and estimation
		Uncertainty
		The problem of estimation, P -values and confidence intervals
		Words of caution on statistical and clinical significance and multiple tests
	On choosing the right statistical analysis method
		The most common statistical methods
		How to decide what method to use?
		Words of caution in the data analysis method selection
	Missingness of data
		Main mechanisms of data missingness
		Main strategies to tackle missing data
		Words of caution for dealing with missing data
	Designing an ophthalmic study
		Study designs, sample size calculation and power analysis
		Words of caution for two eyes: What to do and what not to do?
	Biomarkers
	Ophthalmic imaging data challenges on intersection of statistics and machine learning
	Discussion
	References
Structure-preserving guided retinal image filtering for optic disc analysis
	Introduction
		Optic disc segmentation
		Optic cup segmentation
		Joint optic disc and optic cup segmentation
		Image quality
		Contributions
	Structure-preserving guided retinal image filtering
	Experimental results
		Dataset
		Evaluation metrics
		Results
		Application
			Deep learning-based optic cup segmentation
			Sparse learning-based CDR computation
		Performance on regions with lesions
	Conclusions
	References
Diabetic retinopathy and maculopathy lesions
	Introduction
	The clinical impact of DR and maculopathy lesions
	Type of lesions/clinical features
	Lesion detection and segmentation
		Morphology
		Machine learning
		Region growing
		Thresholding
		Deep learning
		Miscellaneous
		Performance comparison
	Lesion localization
	Conclusions
	References
Drusen and macular degeneration
	Introduction
	Histopathological lesions and clinical classification
		Normal aging of the macula
		Lesions of non-neovascular AMD
		Lesions of neovascular AMD
	Automatic analysis of drusen and AMD-related pathologies
		Drusen detection in retinal fundus photography
			Characterization, classification and quantification of drusen
			Machine learning based approaches
		Drusen segmentation and measurement
		Quantifying drusen area and distinguishing drusen type
			Texture-based methods
		Other imaging modalities
			Angiography
			Scanning laser ophthalmoscopy
			Drusen detection in OCT
		Analysis of other AMD lesions
	Diagnosis of AMD
	Datasets
	Conclusions
	References
OCT fluid detection and quantification
	Introduction
		Intraretinal cystoid fluid
		Subretinal fluid
		Sub-RPE fluid in PED
	OCT fluid quantification
		Segmentation using supervised learning
			Preprocessing and postprocessing
				Denoising
				Retina and layer segmentation
				Data augmentation
			Traditional machine-learning and nonmachine-learning approaches
		Segmentation using weakly supervised and unsupervised learning
		Evaluation
	OCT fluid detection
		Detection using image segmentation
		Detection using image classification
			Traditional machine-learning approaches
		Evaluation
	Clinical applications
		Structure function
		Longitudinal analysis of VA outcomes
			Method
				Obtaining fluid volumes
				Regression model
			Experiments and results
				Dataset
				Regression model
	Discussion and conclusions
	Acknowledgments
	References
Retinal biomarkers and cardiovascular disease: A clinical perspective
	Introduction
	The concept of retinal vascular imaging
	Retinal vascular changes and heart disease
	Retinal vascular changes and stroke
		Clinical stroke
		Subclinical stroke
	Retinal vascular changes and CVD mortality
	Clinical implications
		Retinal vascular imaging as a tool to stratify CVD
		Retinal imaging for clinical trials and outcome monitoring for CVD
	New advances in retinal vascular imaging
		Retinal imaging with artificial intelligence
		Imaging of the choroidal vasculature
		Imaging of the retinal capillary network
		Ultra-widefield retinal imaging
	Conclusions
	References
Vascular biomarkers for diabetes and diabetic retinopathy screening
	Introduction
		The Sino-Dutch collaboration project RetinaCheck
		Vascular analysis-specific biomarkers for early detection and screening
		Layout of this chapter
	Brain- and vision-inspired computing
		The mathematics of V1: Sub-Riemannian geometry in SE (2)
		Orientation scores
		A moving frame of reference
		Sub-Riemannian geometry
		Application: Brain inspired image analysis
	Preprocessing
		Denoising in the SE (2) space
		Vessel segmentation
		Vessel completion
		Validation studies
	Vascular biomarkers
		Vessel width
		Vessel tortuosity
			Single-vessel tortuosity
			Global tortuosity
		SE (2) tortuosity
			Exponential curves in SE(2)
			Fitting the best exponential curve in the orientation scores
			Global tortuosity measurement via the exponential curvature
		Bifurcations
			Murray’s law
			Bifurcation biomarkers
		Fractal dimension
	The processing pipeline
		RHINO software and graphical user interface
	Clinical validation studies
		The Shengjing study
		The Maastricht study
	Discussion
	References
Image analysis tools for assessment of atrophic macular diseases
	The clinical need for automatic image analysis tools in retinal disease
	Overview of analysis tools of atrophic AMD and risk factors for progression to atrophy
	Semiautomated segmentation of atrophic macular diseases
		Heidelberg RegionFinder for atrophic AMD segmentation in FAF images
		Level set approach for atrophic AMD segmentation in OCT and FAF images
	Automated segmentation of atrophic macular diseases
		Supervised classification for atrophic AMD segmentation in FAF images using a traditional machine learning algorithm
		Supervised classification for age-related and juvenile atrophic macular degeneration using an AI deep learning approa ...
	Automated binary classification of OCT risk factors for progression from intermediate AMD to atrophy using an AI deep l ...
	Summary
	Acknowledgments
	References
Artificial intelligence and deep learning in retinal image analysis
	Introduction
	Fundamentals of deep learning
		Fundamentals of neural networks
		Deep convolutional neural networks
		CNNs for semantic image segmentation
	Deep learning applications to retinal disease analysis
		Deep learning for diabetic retinopathy
		Deep learning for age-related macular degeneration
			Deep learning for retinopathy of prematurity and glaucoma
		Deep learning applications in OCT segmentation
	Deep learning for retinal biomarker extraction
		Automatic retinal biomarker discovery
	Datasets
	Conclusion
	References
AI and retinal image analysis at Baidu
	Baidu: Mission, products, and next-steps
		The Baidu mission
		AI in Baidu
		Baidu Brain
			Visual semantic AI
			Speech semantic AI
			Natural language AI
	General architecture of AI retinal image analysis
		Descriptive IQA
			Focus and clarity assessment
			Brightness and contrast assessment
			Illumination evenness assessment
		Disease-specific IQA
		Discussion
		Diabetic retinopathy detection algorithm
			Preprocessing
			Data augmentation
			Classification model
		Glaucoma detection algorithm
		Age-related macular degeneration detection (AMD) algorithm
			Macular AOI location
			End-to-end referable AMD classifier
			Drusen and neovascularization detector
		Interpretation module
		Experimental results and real-world application
			Image quality assessment
			Diabetic retinopathy
			Glaucoma
			Age-related macular degeneration
			Real-world application
		Outlook of Baidu retina system
		Acknowledgments
	References
The challenges of assembling, maintaining and making available large data sets of clinical data for research
	Introduction
	Sources of images and associated data
		Research collected images
		Routinely collected images
		Sources of ground truth data
		Linking clinical data to imaging data
	Data governance
		Key data protection terminology and concepts
		Applications to access data for research
	Controls
		Safe data
			Identifying information
			Acceptance threshold for re-identification
			Transformation of data
			Considerations when anonymizing pixel data
			Software to anonymize DICOM images
		Safe people and organizations
			Indexing and linking
			Trusted third parties
			Who will be accessing the research data
		Safe access
			Transferring data
			Data hosted on a researcher managed environment
			Safe Havens/trusted research environments
			Federated or distributed analysis
			Challenges of assembling large quantities of clinical data within data governance controls
	Conclusions
	References
Technical and clinical challenges of A.I. in retinal image analysis
	Introduction
	Progression of A.I. in retinal imaging
	Technical challenges
		Quantity of data
		Quality of data
		Heterogeneous data
		Unbalanced data
		Incomplete data
		Private data
		Model generalizability
		Model interpretability
		Model maintainability
		Model deployability
	Clinical challenges
		Variation in DR classification systems and reference standards
		Disagreement in clinical ground truth
		Integration into clinical workflows
		Privacy and data collection
		Assignment of liability
		Patient and physician acceptance of “black box” models
		Expectation management
	Conclusion
	References
Index
	A
	B
	C
	D
	E
	F
	G
	H
	I
	J
	K
	L
	M
	N
	O
	P
	Q
	R
	S
	T
	U
	V
	W
	Y
	Z