Quantitative Analysis Method for Bacterial Cells in SEM Image using Deep Learning

Anonymous

TL;DR: We describe a method to count each type of bacteria in SEM image using deep learning segmentation and classification.

Abstract: In this paper we propose a method to segment, classify and quantitatively analyze bacteria from a given Scanning Electron Microscope (SEM) image of the bacterial sample. Thousands of bacteria lives in the human gut and recent studies have shown that the quantitative features of the microbiome, such as co-existence ratio of different bacteria, can be indicative of the health condition in humans. Therefore, to realize a system to quantitatively analyze the gut bacteria of humans, we propose a method to segment, classify and calculate the ratio of the bacteria contents for a few well-known bacteria types. Our method achieves more than 90% recall for all of original three datasets. Additionally, we also introduce a novel image processing based touching object separation algorithm which is applied within the framework of our system. Subsequently, we show the comparison results between another state-of-the-art segmentation method and the introduced algorithm and we empirically report that our new algorithm has a better performance.

Keywords: bacteria, SEM, segmentation, classification, neural network