The specific goal of this proposal is to use adaptive capabilities and parallel computational properties of novel artificial neural networks (ANNs) for automatic classification of mouse chromosomes. The overall objective is to provide a computer software system to automate the mouse karyotyping process, thus signifcantly reducing the time and human effort neccessary, and improving the quality and efficiency of analysis. Analysis of mouse chormosomes is important to many fields of genetic research. Example applications include gene mapping by in situ hybridization to nucleic acid probes to mouse chromosomes, substance screening which requires chromosome analysis of large numbers of animals (most drequently mice) after treatment with the substance, and testing breeders to maintain stocks of mice that carry chromosome aberrations of interest. Mouse chromosomes are significantly more difficult to classify than human chromosomes. Although automated human karyotyping systems are available, none have been used successfully to classify mouse chromosomes automatically. Artificial Neural Networks (ANNs) represent an emerging technology rooted in many disciplines. An ANN is an information processing system that has certain performance characteristics in common with biological neural networks. ANNs have been developed as generalizations of mathematical models of human cognition or neural biology. An ANN consist of a network architecture and a learning paradigm. The architecture defines the topology and complexity of the network necessary to acquire a specific level of knowledge in the area of interest, which in this case is recognition of specific chromosomes. The learning paradigm involves a training process for embedding the knowledge in the network. We propose to design, train, test, and eveluate two novel ANNs, Radial Basis Function (RBF) and Probabilistic Neural Network (PNN), for classification of mouse chromosomes. The superior ANN will be integrated into an existing karyotyping software distribution and will replace the conventional classification module. Use of the existing karyotyping software allows us to concentrate on the important issue of classification by providing the tools for capture of metaphase spreads under a light microscope , segmentatiion and enhancement of digitized images, as well as a user interface. To achieve the objective of this porposal, the specific plans are to 1. obtain raw images of mouse chromosomes 2. identify distinctive features of mouse chromosomes 3. develop the programs for automatic extraction of the features, 4. prepare the training and testing data sets, 5. design, train, and test the radial basis function (RBF) neural network classifier, 6. design, train, and test the probabilistic neural network (PNN) classifier, 7. select the optimal artificial neural network classifier for mouse chromosomes, 8. improve the classification performance, 9. develop the mouse karyotyping qystem.
