University of Groningen Brain-inspired computer vision with applications to pattern recognition and computer-aided diagnosis of glaucoma Guo, Jiapan IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from it. Please check the document version below. Document Version Publisher's PDF, also known as Version of record Publication date: 2017 Link to publication in University of Groningen/UMCG research database Citation for published version (APA): Guo, J. (2017). Brain-inspired computer vision with applications to pattern recognition and computer-aided diagnosis of glaucoma. [Groningen]: University of Groningen. Copyright Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons). Take-down policy If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim. Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum. Download date: 18-03-2019
Brain-inspired Computer Vision with Applications to Pattern Recognition and Computer-aided Diagnosis of Glaucoma Jiapan Guo
The research work described in this thesis was carried out at the Intelligent Systems group of the Johann Bernoulli Institute for Mathematics and Computer Science, University of Groningen, The Netherlands. This work was financially supported by the China Scholarship Council (CSC). ISBN: 978-94-034-0274-1 (printed version) ISBN: 978-94-034-0273-4 (digital version)
Brain-inspired Computer Vision with Applications to Pattern Recognition and Computer-aided Diagnosis of Glaucoma PhD thesis to obtain the degree of PhD at the University of Groningen on the authority of the Rector Magnificus Prof. E. Sterken and in accordance with the decision by the College of Deans. This thesis will be defended in public on Monday 04 December at 11.00 hours by Jiapan Guo born on 18 December 1986 in Tsukuba, Japan
Supervisor Prof. dr. N. Petkov Co-supervisor Dr. G. Azzopardi Assessment Committee Prof. dr. X. Jiang Prof. dr. D. P. Mandic Prof. dr. A. C. Telea
to my happy and lovely family
Contents List of Figures List of Tables Acknowledgements iv vi vii 1 Introduction 1 1.1 Scope..................................... 3 1.2 Thesis organization............................. 6 2 Inhibition-augmented Trainable COSFIRE Filters for Shape Detection and Object Recognition 9 2.1 Introduction................................. 10 2.2 Method.................................... 13 2.2.1 Overview.............................. 13 2.2.2 Gabor Filters............................ 15 2.2.3 Configuration of an Inhibition-augmented COSFIRE Filter.. 15 2.2.4 Configuration with Multiple Negative Prototypes....... 18 2.2.5 Implementation........................... 21 2.2.6 Tolerance to Geometric Transformations............. 26 2.3 Applications................................. 26 2.3.1 Detection of Retinal Vascular Bifurcations............ 26 2.3.2 Detection of Architectural and Electrical Symbols....... 31 2.3.3 Recognition of Handwritten Digits................ 39 2.4 Discussion.................................. 41 2.5 Conclusions................................. 43 3 Automated Optic Disc Localization and Diameter Estimation in Retinal Fundus Images 45 3.1 Introduction................................. 46 i
CONTENTS 3.2 Detection of the Optic Disc by Circular Hough Transform....... 48 3.2.1 Preprocessing............................ 48 3.2.2 Vessel Segmentation and Elimination.............. 49 3.2.3 Edge Detection by Gabor Filters................. 51 3.2.4 Localization of the Optic Disc................... 51 3.2.5 Experimental Results........................ 52 3.3 Detection of the Optic Disc by Trainable COSFIRE Filters....... 53 3.3.1 COSFIRE Filters........................... 53 3.3.2 Configuration of a Vasculature-selective COSFIRE Filter... 55 3.3.3 Response of a Vasculature-selective COSFIRE Filter...... 56 3.3.4 Localization of the Optic Disc................... 59 3.3.5 Delineation of the Optic Disc Boundary............. 60 3.3.6 Implementation and Experimental Results........... 61 3.3.7 Experimental Results........................ 68 3.4 Discussion.................................. 70 3.5 Conclusions................................. 72 4 Automatic analysis of retinal fundus images for glaucoma screening based on vertical cup-to-disc ratio 75 4.1 Introduction................................. 77 4.2 Proposed method.............................. 81 4.2.1 Overview.............................. 81 4.2.2 Localization and Delineation of the Optic Disc......... 83 4.2.3 Segmentation of the Cup by K-means Clustering....... 83 4.2.4 Vertical Cup-to-disc Ratio (VCDR)................ 85 4.3 Implementation and experimental results................ 86 4.3.1 Data sets and the manual annotation from the ophthalmologist 86 4.3.2 Edge Preserving Smoothing.................... 86 4.3.3 Implementation of the proposed approach........... 87 4.3.4 Experimental Results........................ 88 4.4 Discussion.................................. 92 4.5 Conclusions................................. 97 5 Summary and Outlook 99 5.1 Summary................................... 99 5.2 Outlook.................................... 101 Samenvatting 103 Bibliography 107 ii
CONTENTS Research Activities 119 iii
List of Figures 1.1 Example of the game spot the difference............... 1 1.2 Example of a bifurcation and a crossover................ 3 1.3 Example of a healthy and a suspicious retina.............. 6 2.1 Examples of pairs of patterns....................... 10 2.2 Selectivity of a TEO neuron........................ 12 2.3 Example of a positive and a negative prototypes............ 14 2.4 Gabor filter and its responses....................... 15 2.5 Configuration of COSFIRE filters by two different prototypes.... 17 2.6 The structure of an inhibition-augmented filter............. 19 2.7 Multiple negative prototypes and the resulting filter.......... 20 2.8 Application illustration of an inhibition-augmented COSFIRE filter. 23 2.9 Illustration of the tolerances........................ 24 2.10 The responses of the configured filter................... 25 2.11 Example of a retinal fundus image and the vascular bifurcations... 27 2.12 Examples of positive and negative prototype patterns.......... 28 2.13 A set of four bifurcations taken from the DRIVE data set....... 29 2.14 Precision-recall plots............................ 30 2.15 Symbols in the GREC 2011 data set.................... 32 2.16 Examples of noisy images......................... 32 2.17 Examples of architectural symbols.................... 33 2.18 Structures of the configured COSFIRE filters.............. 33 2.19 The harmonic mean............................. 34 2.20 Confusion matrices............................. 35 2.21 An example image of an electrical circuit................ 37 2.22 Examples of MNIST data set........................ 39 2.23 Recognition rate plots on the MNIST data set.............. 40 3.1 Example of a retinal fundus image.................... 46 3.2 Illustration of the proposed method................... 49 3.3 Sketch of a B-COSFIRE filter........................ 50 3.4 Illustration of the disc detection result by CHT............. 52 3.5 Examples of determined optic discs................... 54 iv
LIST OF FIGURES 3.6 Example of a retinal fundus image.................... 56 3.7 Major vessel course............................. 57 3.8 Configuration of a vessel-selective COSFIRE filter........... 57 3.9 Application of the vascular-selective filter................ 58 3.10 Configuration of the disc-selective filter................. 60 3.11 Application of the disc-selective filter.................. 61 3.12 Optic disc boundary detection....................... 62 3.13 Examples of fundus images from different data sets.......... 63 3.14 Examples of vasculature-selective COSFIRE filters........... 66 3.15 Examples of disc-selective filters..................... 67 3.16 Examples of the disc delineation results................. 68 3.17 Examples of incorrect localization of the optic disc........... 71 4.1 Illustration of the eye anatomy...................... 78 4.2 Example of a retinal fundus image.................... 79 4.3 Example of a retina with pathology.................... 80 4.4 Flowchart of the method.......................... 82 4.5 Initialization of K-means clustering for cup segmentation....... 84 4.6 Cup segmentation result by the K-means clustering.......... 85 4.7 VCDR computation............................. 85 4.8 Examples of manual annotations..................... 87 4.9 Box-whisker plot of the manual VCDRs................. 91 4.10 Box-whisker plot of the VCDR...................... 92 4.11 Distribution of the VCDRs on reliable images.............. 93 4.12 Bland-Altman plot............................. 94 4.13 ROC curves................................. 95 4.14 Examples of unreliable images...................... 96 4.15 Example of unobvious cup excavation.................. 96 v
List of Tables 2.1 The optimal values of η and t 3....................... 30 2.2 Recognition rates.............................. 36 2.3 Comparison of the results with other methods............. 38 3.1 List of public available data sets...................... 63 3.2 Localization accuracy on all images.................... 69 3.3 Results of the optic disc segmentation on all images.......... 70 4.1 Localization performance......................... 89 4.2 Disc segmentation performance...................... 89 4.3 Cup segmentation performance...................... 90 vi