Segmentation of Brain CT Image Machine Learning.
- Author:
Ho Sung KIM
1
;
Young Ran LEE
Author Information
1. Department of Computer Science Sungshin Women's University, Korea.
- Publication Type:Original Article
- Keywords:
medical image;
segmentation;
region growing;
rule;
learning
- MeSH:
Machine Learning*;
Brain*;
Diagnosis;
Knowledge Bases;
Learning;
Noise
- From:Journal of Korean Society of Medical Informatics
1997;3(2):193-199
- CountryRepublic of Korea
- Language:Korean
-
Abstract:
A medical image segmentation is the primary issue in computer aided diagnosis. The traditional methods did not perform the image segmentation well because of varieties of image, inadequate informations, noises, uncertain images, and deficient image data. We Propose a new medical image segmentation by machine learning using background knowledge of segmentation pattern. The proposed algorithm is applied to real brain CT images. First, a region growing algorithm extracts the regions and statistical data. Also, shape informations about each regions are gathered. A supervisor makes a set of learning examples by selecting the regions which should be in one region. In the next step, some rules for merging regions are discovered from common properties of the examples. Also there will be verification procedure whether the pattern is the desired one. The procedure is achieved by machine learning technique from the patterns of positive or negative examples. The systems try to recognize the improved patterns in the next step, and make a knowledge base for the segmentation. From the experimental results of the proposed algorithm which is applied to various brain images, we obtain an adaptable knowledge base and a segmented image with proper regions of brain shape.
