Knowledge Representation Using Case Based Reasoning

Case-based reasoning (CBR) approaches have been used extensively as a means of directly utilizing image properties. CBR is an approach to computer-based cognition that involves reasoning from prior experiences. It solves new problems by adapting solutions that were used to solve old problems. The knowledge base of a CBR system consists of cases indexed by their pertinent features.

Perner [16] proposes a novel method of image segmentation based on CBR. The CBR unit for image segmentation consists of a case base in which formerly processed cases are stored. Accompanying each case is information regarding the parameters used in the segmentation of the image. On test, a similarity measure is used to select the most similar case in the archive from which the segmentation parameters are selected and used to segment the test image. The author hypothesizes that images having similar characteristics will show good segmentation results when the same segmentation parameters are applied. The image information comprises a vector of statistical features extracted from the gray scale histogram of the matching image. In the author's implementation, nonimage information is obtained from CT image headers, such as sex, age, CT-slice thickness, etc. The final similarity measure comprises both image and nonimage information. The segmentation of the image is performed using histogram analysis. The parameters for this process comprise those defining a smoothing function for the histogram and a set of thresholds used for histogram analysis. The system has been evaluated using a 130 image case base and by comparing the automatic CBR segmentation with that of an expert clinician on 600 CT images of the brain. The author reports a linear correlation coefficient of 0.85 between the CBR segmentation and those manually drawn by an expert clinician.

0 0

Post a comment