Biopolymer segmentation from CLSM microscopy images using a convolutional neural network

Abstract

AbstractConfocal microscopy allows visualization of biopolymer networks at the nano scale. Analyzing the structure and assembly of protein networks from images requires a segmentation process. This has proven to be challenging due to multiple possible sources of noise in images as well as exhibition of out‐of‐focus planes. Here, we present a deep learning‐based segmentation procedure for confocal laser scanning microscopy images of biopolymer networks. Utilizing an encoder‐decoder network architecture, our deep neural network achieved a dice score of 0.88 in segmenting images of filamentous temperature sensitive Z proteins from chloroplasts of Physcomitrella patens, a moss.