Feature Augmented Deep Neural Networks for Segmentation of Cells
In this work, we use a fully convolutional neural network for microscopy cell image segmentation. Rather than designing the network from scratch, we modify an existing network to suit our dataset. We show that improved cell segmentation can be obtained by
- PDF / 2,899,051 Bytes
- 13 Pages / 439.37 x 666.142 pts Page_size
- 116 Downloads / 214 Views
Department of Information Technology, Uppsala University, Uppsala, Sweden [email protected] 2 SciLifeLab, Uppsala, Sweden
Abstract. In this work, we use a fully convolutional neural network for microscopy cell image segmentation. Rather than designing the network from scratch, we modify an existing network to suit our dataset. We show that improved cell segmentation can be obtained by augmenting the raw images with specialized feature maps such as eigen value of Hessian and wavelet filtered images, for training our network. We also show modality transfer learning, by training a network on phase contrast images and testing on fluorescent images. Finally we show that our network is able to segment irregularly shaped cells. We evaluate the performance of our methods on three datasets consisting of phase contrast, fluorescent and bright-field images. Keywords: Deep neural network · Feature augmentation · Cell segmentation · Convolutional neural network · Unstained cells
1
Introduction
Observation of biological samples over prolonged periods of time is commonly used to study phenotypical changes due to variations in environmental conditions or genetic modifications. High-throughput high-content screening is used to analyse many biological processes simultaneously [1]. It is tedious for human observers to monitor changes at the cellular level over long time. Automated image analysis algorithms are widely used to simplify and quantify the analysis process [2]. For automated image analysis at the cellular level, a commonly used approach is to segment the cellular regions and track the cell segments over time [3]. The cell segmentation is a crucial step in this process, which affects the quality of the cell tracking results. In this work, we aim to segment cells in timelapse microscopy image sequences. Cell segmentation is a challenging process, especially when the cells are unstained. Deep Neural Networks (DNN) using Fully Convolutional Neural Networks (FCNN) have shown excellent results in semantic segmentation [4]. FCNNs were also used in segmenting unstained cells in microscopy images [5,6]. The network structures of these high performing FCNNs, as opposed to traditional DNNs [7,8], suggest that designing the proper c Springer International Publishing Switzerland 2016 G. Hua and H. J´ egou (Eds.): ECCV 2016 Workshops, Part I, LNCS 9913, pp. 231–243, 2016. DOI: 10.1007/978-3-319-46604-0 17
232
S.K. Sadanandan et al.
(a) EFD
(b) EPD
(c) MBD
Fig. 1. (a) Input E. coli fluorescent dataset (EFD) (b) E. coli phase contrast dataset (EPD) and (c) mouse mammary cells bright-field dataset (MBD).
deep network is non-trivial. Often, a network that gives good results on a particular dataset may not give good results on another dataset. Fusing features from different layers of deep networks [9] or combining deep features with hand-crafted features [10] were used in video action recognition tasks, where the authors used ‘late fusion’, i.e., they combined features at later layers of the deep network for classification. A combination
Data Loading...
