Classification of Giemsa-stained human chromosomes using multi-layer neural network

The classification of Giemsa-stained human chromosomes using a multi-layered neural network was examined for the image data of 460 chromosomes. Features extracted from the normalized density profile were used as input vectors for training of the neural network. Two learning sets, each consisting of 27 and 52 feature vectors, have been prepared for the training and recall phase. The experiments were carried out with various numbers of processing elements (PEs) in a hidden layer to determine the optimal number of PEs under given conditions. The results show that multi-layered neural networks have the potential for classifying Giemsa-stained human chromosomes