Efficient neural network classifier of Medaka embryo using morphological pattern spectrum

In this paper, an efficient neural network classification system for mutant embryos of Medaka using morphological pattern spectrum is proposed. The Medaka, Oryzias Iatipes is a vertebrade experimental model system developed in Japan, whose live eggs are transparent allowing observation of body pattering before birth. Millions of mutant embryos were generated by randomly inducing mutations into Medaka genome and have screened for defects in body pattering. Analysis of these mutations will be important information for the therapeutic purpose of human. First, the various situations of individual embryos in color microscopic image are represented as binary image by segmentation process with histogram and morphological filters. Next, computationally efficient morphological pattern spectrum is calculated. The selected relative pattern spectrum which is location, rotation and scale invariant can be a feature vector of the object discrimination. Further, using the relative pattern spectrum as input signal pattern, a multi-layer feed forward neural network is applied to classify the mutant. In the simulations, the three-layer neural network is trained using learning pattern spectrums. It is shown that the method can automatically classify the defects of mutant embryos.