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

• DOI: 10.3389/fcomp.2021.613981
• Publication Date: 2021-05-13T08:54:30Z
• Authors: Tran Minh Quan; David Grant Colbu...; Won-Ki Jeong
• Keywords: Connectomics

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