Chromosome Classification Based on Wavelet Neural Network

Abstract

Karyotyping, manual chromosome classification is a difficult and time consuming process. Many automated classifiers have been developed to overcome this problem. These classifiers either have high classification accuracy or high training speed. This paper proposes a classifier that performs well in both areas based on wavelet neural network (WNN), combining the wavelet into neural network for classification of chromosomes in group E (chromosomes 16, 17 and 18). The nonlinear characteristic of the network which is derived from wavelet specification improves the training speed and accuracy of the nonlinear chromosome classification. The network inputs are nine dimensional feature space extracted from the chromosome images and the outputs are three classes. The simulation result on the chromosomes in the Laboratory of Biomedical Imaging shows that the success rate of WNN was 0.93%, that is comparable to the traditional neural network (ANN) with 0.85% success rate. The number of iterations for training to reach 0.04% error rate is only 200 where it is 3500 iterations for ANN. According to the experimental results WNN achieves high accuracy with minimum training time, which makes it suitable for real-time chromosome classification in the laboratory.