ABSTRACT
To boost the performance of level set algorithms, we propose the A-Levelset algorithm, which cascades the level set and active shape model (ASM). The A-Levelset-based ARGALI system is built to automatically segment the optic cup and optic disc from 2D digital fundus images. The ARGALI system further calculates the cup-to-disc ratio (CDR), which is an important indicator in glaucoma assessment and diagnosis. The ARGALI system was tested on a large clinical image collection of 2616 patients in order to estimate the CDR
CONTENTS
7.1 Introduction of Glaucoma Diagnosis .............................................................................. 130 7.1.1 Glaucoma Is an Irreversible Blinding Optic Neuropathy ................................ 130 7.1.2 Early Detection of Glaucoma Is Important ........................................................ 130 7.1.3 Glaucoma Is the Leading Cause of Irreversible Blindness .............................. 130 7.1.4 Currently There Are No Effective Tools for Glaucoma Screening ................. 131 7.1.5 Medical Cost of Glaucoma Management Is Substantial .................................. 131 7.1.6 A Sensitive, Reproducible, and Effective Tool for Screening
Glaucoma Is Needed ............................................................................................. 131 7.1.7 Glaucoma Is Highly Prevalent, Particularly in Asia ......................................... 132 7.1.8 The Economic Cost and Societal Burdens of Glaucoma Are Substantial ...... 132
7.2 A-Levelset Algorithm and the ARGALI System ........................................................... 133 7.2.1 ASM and Level Set for Fundus Image Segmentation ....................................... 133 7.2.2 A-Levelset Algorithm ............................................................................................ 133 7.2.3 ARGALI System for Fundus Image Segmentation ........................................... 134 7.2.4 The First Step in ARGALI: Level Set-Based Optic Cup Segmentation .......... 135 7.2.5 The Second Step in ARGALI: Register the Level Set Segmented Cup
with the ASM Mean Shape ................................................................................... 135 7.2.6 The Third Step in ARGALI: Run ASM with the New Registered
Mean Shape as the Initial Contour ...................................................................... 136 7.3 Results ................................................................................................................................. 137
7.3.1 Data Sets Used for ARGALI System Evaluation ................................................ 137 7.3.2 ARGALI Experimental Results and Analysis .................................................... 137
7.4 Conclusions ......................................................................................................................... 139 References ..................................................................................................................................... 139
values. The extensive experimental results clearly show that ARGALI outperforms the level set-based approach by reducing the mean absolute error rate of CDR measurement from 0.349 to 0.21 and the mean square error rate from 0.156 to 0.07. ARGALI demonstrates for the first time the capability of automatic CDR measurement in a large clinical data set. It paves the way for automatic objective glaucoma diagnosis and screening using widely available fundus images.
