Iterative Contextual Recurrent Classification of Chromosomes

摘要

Recurrent connectionist models provide a method to represent dynamic patterns in a neural network. In this work we present a method for chromosome classification based on an almost unexplored neural network technique for this task. A partially recurrent connectionist model, the Elman network, is managed to capture the dark and light band patterns of the different classes. The proposed method is completed with the formulation of the ICC (iterative contextual classification) algorithm in order to restrict the classification to the cell context, and is applied to the neural network results. The Copenhagen data set was used in the experiments, where a cross-validation method was applied in order to obtain statistically representative results using the complete corpus. The entire system obtained very good results for this task, improving the performance of other neural network approaches.