FusionNet: A Deep Fully Residual Convolutional Neural Network for Image Segmentation in Connectomics

Abstract

Cellular-resolution connectomics is an ambitious research direction with the goal of generating comprehensive brain connectivity maps using high-throughput, nano-scale electron microscopy. One of the main challenges in connectomics research is developing scalable image analysis algorithms that require minimal user intervention. Deep learning has provided exceptional performance in image classification tasks in computer vision, leading to a recent explosion in popularity. Similarly, its application to connectomic analyses holds great promise. Here, we introduce a deep neural network architecture, FusionNet, with a focus on its application to accomplish automatic segmentation of neuronal structures in connectomics data. FusionNet combines recent advances in machine learning, such as semantic segmentation and residual neural networks, with summation-based skip connections. This results in a much deeper network architecture and improves segmentation accuracy. We demonstrate the performance of the proposed method by comparing it with several other popular electron microscopy segmentation methods. We further illustrate its flexibility through segmentation results for two different tasks: cell membrane segmentation and cell nucleus segmentation.

A Novel Deep Structure U-Net for Sea-Land Segmentation in Remote Sensing Images

“…Moreover, for the proper reconstruction, a single identity mapping [44] in the dense block has been used to merge the feature maps which are generated from the convolution layers. Hence, these novel components significantly improved the overall segmentation results and robustness of the RDU-Net as compared to U-Net [9], DeepUNet [5], and FusionNet [15].…”

“…1 {y (i,j) = q} is the stating function, where, if the ground truth of the consistent pixel is q, then one is obtained; otherwise, zero is obtained. For the regularization task, To evaluate the efficiency of RDU-Net, we compare it with the U-Net [9], FusionNet [15] and the DeepUNet [5] with the same datasets and experimental settings. The network structures of the mentioned models are either downloaded directly from the GitHub web pages or implemented with the technical details that the authors provided.…”

“…We intend to reduce the distance between the layers during the backpropagation that has a significant role in reducing the computation costs. Meanwhile, short-range connections and identity mapping [44] are combined to build a more effective network than the current approaches [5], [15], where identity mapping is a fundamental aspect of learning in deep networks and has significant effects in the training of the networks [24].…”

“…The proposed method is based on U-Net which contains densely connected residual network blocks that significantly improve the overall results while having low computation costs. • To assess the performance of the proposed model, we perform a wide range of comparisons between U-Net [9], FusionNet [15], DeepUNet [5], and RDU-Net. The experiments show that RDU-Net is far more efficient than the other state-of-the-art networks.…”

A Deep Learning Framework for Segmenting Brain Tumors Using MRI and Synthetically Generated CT Images

“…The final network was used as post-processing to enhance the tumor region. They used three modified versions of the FusionNet [15] network architecture in the Caffe [16] platform. The three networks were identical except for their final segmentation layers.…”