Automated extraction and quantification of macular drusen from fundal photographs

Princess Margaret Hospital, Perth
Australian and New Zealand Journal of Ophthalmology 11/1994; 22(1):7 - 12. DOI: 10.1111/j.1442-9071.1994.tb01688.x


The objective quantification of drusen (and other macular lesions) should have applications epidemi-ologically, in the study of the natural history of drusen, and with such instruments as the scanning laser ophthalmoscope. The automated extraction of drusen from photographs is technically difficult because of uneven macular reflectance, and the confusing pattern of darker vessels. We have developed a method using an IBM personal computer, an image digitising board and specially written software. Once the image is digitised, no further input from the operator is necessary. We present the results of manual counting versus automated counting on a small series of patients with drusen. The automated technique is highly reproducible, and will calculate the retinal area occupied by drusen. The area and numbers of drusen can be compared over time, giving an index of progression. Hard drusen are fairly well detected, but the detection of soft drusen with their lower contrast remains a problem. The technique cannot distinguish between drusen and other pale lesions (e.g., atrophic retinal changes).

6 Reads
  • Source
    • "Several studies on automatic detection of drusen have been published in the last twentyyears . The majority of works were based on applying image thresholds to separate drusen from background or other structures [2] [3] [4] [5] [6] [7] [8] [9]. We have also contributed to this subject by introducing new image processing algorithms for illumination correction [10] and for drusen detection and modeling [11] [12]. "
    [Show abstract] [Hide abstract]
    ABSTRACT: The assessment of techniques of automatic detection of structures in retinal images is mostly done through the comparison between the results produced automatically and the ones produced manually by specialists. When the analyses are subjective some disparities are common to appear among different specialists as well as within the repeated analysis of one specialist. A decisive mechanism is therefore needed to obtain more accurate results. In this article it is presented the weighed matching analysis method, which was developed to be a pixel to pixel analysis that uses the statistical significance of the observations to differentiate positive and negative pixels. It is based on the creation of a probabilities map, which results from the specialists’ markings, followed by the calculus of sensitivity, specificity and kappa coefficient between one analysis and the probabilities map. This method was validated with a dataset having 22 retinal images with visible drusen. These were marked by 8 independent specialists and by the automatic detection method. The results of sensitivity, specificity and kappa coefficient were calculated using both the weighed matching analysis method and the binary method, for comparison purposes. It was concluded that this method improved the binary matching analysis, especially on image sets that contain analysis with significant variability, by automatically removing outlier pixels and by having rewards and penalizations with different weights based on the probabilities map values when there is no absolute agreement. Also, this method allows not only the method’s validation, but also the quantitative comparison between specialists to identify outliers.
    IBERSENSOR - 7TH Ibero-American Congress on Sensors; 01/2010
  • Source
    • "Uneven illumination artifacts of the retina forced the application of sophisticated adaptive thesholding utilizing Otsu's method, e.g. Morgan et al. and others [5]. Interactive techniques have been applied, driven by limitations of automated segmentation and algorithms. "
    [Show abstract] [Hide abstract]
    ABSTRACT: The purpose of this paper is to present a novel approach for extracting image-based features for classifying age-related macular degeneration (AMD) in digital retinal images. 100 retinal images were classified by an ophthalmologist into 12 categories based on the visual characteristics of the disease. Independent Component Analysis (ICA) was used to extract features at different spatial scales to be used as input to a classifier. The classification used a type of regression, partial least squares. In this experiment ICA replicated the ophthalmologist's visual classification by correctly assigning all 12 images from two of the classes.
    Image Analysis and Interpretation, 2008. SSIAI 2008. IEEE Southwest Symposium on; 04/2008
  • Source
    • "Both algorithms achieved similar results on images from the same patient but they produced a considerable number of false positive detections. In 1994 Morgan et al. [6] from University of Western Australia published a paper where it was proposed an automatic Drusen quantification based on the application of the Otsu threshold to local windows of 16x16 pixels. The results obtained with the automatic quantification were compared with manual counting done by three experts. "

    MEDSIP-2004; 01/2004
Show more

Similar Publications