Operations Useful for Similarity-Invariant Pattern Recognition
Journal of the ACM (JACM)
An introduction to support Vector Machines: and other kernel-based learning methods
An introduction to support Vector Machines: and other kernel-based learning methods
A Max-Min Measure for Image Texture Analysis
IEEE Transactions on Computers
Computer Methods and Programs in Biomedicine
Automatic measurement of vertical cup-to-disc ratio on retinal fundus images
ICMB'10 Proceedings of the Second international conference on Medical Biometrics
| Hi-index | 0.00 |
This paper describes automated lesion detection in retinal images. Physicians and ophthalmologists assess retinal images for several kinds of lesions, including hemorrhages, exudates, and arteriolar narrowing. Hemorrhage is a major sign of diabetic retinopathy, which is the second most common cause of vision loss. Arteriolar narrowing is a major sign of hypertensive retinopathy. The aim of this study was to measure arteriolar-to-venular diameter ratio for the detection of arteriolar narrowing and to develop a hemorrhage detection method. Blood vessels and hemorrhages were extracted using a double-ring filter. This filter device calculates the difference between the average pixel values of the inside and outside regions. Arteriolar narrowing is determined based on major arteriolar-to-venular diameter ratios. Thus, the major blood vessels were extracted and the arteriolar-to-venular diameter ratio was automatically calculated based on the artery and vein diameter measurements. Finally, the hemorrhage candidates remained after the blood vessels were "erased" from the image and hemorrhages were detected by machine learning methods using 64 texture features. We tested 20 retinal images from the DRIVE database to evaluate our proposed arteriolar-to-venular diameter ratio measurement method. Both the average error and the standard deviation of the arteriolar-to-venular diameter ratio measurements were 0.07 ± 0.06. We evaluated the proposed method for hemorrhage detection by testing 71 retinal images, including 53 images with hemorrhages and 18 normal ones. The sensitivity and specificity for the detection of abnormal cases were 83% and 67%, respectively.