Implementation of Deep Neural Networks for Medical Image Segmentation

Abstract

Image segmentation is a powerful tool for image analysis that has applications in autonomous vehicles, face recognition and healthcare. The basic functionality is to transform a given image into something comprising of more meaning and clarity. Segmentation has been extensively used for the analysis of cell images, which play an important role in tracking disease progression and identifying pathologies. To address the problem of cell segmentation, a popular encoder-decoder based convolution network in the literature has been increasingly used, that is known as UNet. Its architecture has been updated by using residual networks (ResNet) in the encoder part, resulting into a deeper segmentation network which leads to an improved accuracy. In this research, we have used ResNet with identity mappings as the encoder in U-Net and its performance is evaluated on the basis of comparison with previously used encoder-decoder based networks. After training the networks, it was seen that the proposed network outperformed U-Net and U-Net with ResNet encoder in terms of the dice score and IoU score. The results produced in this work can be used as a baseline for medical image segmentation. Keywords: Image Segmentation, Cell Images, U-Net. x