The Detection of Opacification on the Posterior Capsule by Texture-based Segmentation


London University (September, 1995)

Mark Holden,

Department of Physics, Wheatstone Laboratory, King's College London, Strand, London WC2R 2LS

e-mail: mark.holden@kcl.ac.uk

Abstract

As part of a study on the cataracts eye disorder investigations on how to automatically quantify the extent of the associated opacification were carried out. Some 169 digital images of the eyes of patient's fitted with an artificial lens were systematically analysed. Observations made of these images led to the identification of four distinct textural region classes. From an analysis of the differences in textural properties of these region classes a set of feature extractors were proposed and validated experimentally. A feature vector extractor algorithm was developed that from a set of example sub-image region classes generated a set of patterns in 4D feature space. The patterns were then classified with a multi-layer perceptron back propagation neural network. Preliminary results on data not used in training indicated a 100 percent classification accuracy for 800 pixels extracted from clear regions and a 99 percent classification accuracy for 800 pixels extracted from opacified ones.

  • Dissertation in pdf format
    Mark Holden Radiological Sciences.
    Email: mark.holden@kcl.ac.uk Last modified: 4 May, 2001.