Automatic White Blood Cell Classification Using Biased-Output Neural Networks with Morphological Features
Nipon Theera‐Umpon · Thammasat International Journal of Science and Technology · 2003
Numbers of white blood cells in different classes help doctors to diagnose patients. A new set of features based on the mathematical morphology in the white blood cell classification problem is proposed in this paper. The proposed features are the maximum value of a pattern spectrum, the iocation where the maximum value of a pattern spectrum occurs, the first and second granulometric moments. We also propose a method to unbias neural networks by biasing the desired output using a priori information of the number of samples in each class. Regular artificial neural networks and the biased-output neural networks are applied in the experiments using the five-fold cross validation as the testing method. The results show the good performances of classifiers using our biased-output neural networks and our proposed morphology-based features.