Human blastocyst segmentation using neural network

In this paper, a new method to segment the blastocyst images in JPEG compressed domain is proposed. We exploit valuable features of a DCT transform to automate the segmentation process in a blastocyst image. A two layer feedforward backpropagation neural network is trained using the derived features of DCT coefficients of JPEG images to learn the characteristics of blastocyst different components. The precision value for the identification of ZP, TE and ICM detection in test data are 0.80, 0.69 and 0.76, while the recall values are 0.88, 0.78 and 0.87, respectively.