Image segmentation based on U-Net plus plus ?network method to identify Bacillus Subtilis cells in micro-droplets

The study of the formation of Bacillus subtilis biofilms in microdroplets has great significance for understanding the biofilms growth in extreme environments. Due to the cell motion, cells observed by microscope have the characteristics of fuzzy edge information, low contrast and noise. it is difficult to segment targets from the background by traditional segmentation methods, but artificial intelligence has a better performance in the field of biological images. In the experiment, a two-stage cross microfluidic tube control system is used to obtain mono-disperse droplets containing Bacillus subtilis, and the image data are captured by a fast camera through a dark field microscope. In this paper, U-Net++ neural network model is used to identify cells. The encoder is used to extract high-level features of images. The decoder restores the features extracted by the encoder. Dense jump connection reduces the semantic gap between encoder and decoder, captures details and improves segmentation performance. Compared with the traditional segmentation methods, the U-Net++ model can be applied even in cases of low contrast and noise, and improves the accuracy and robustness of image segmentation. The U-Net++ method is compared with the traditional threshold segmentation method in terms of a series of metrics (accuracy, recall, F1 score, Intersection over Union). It is demonstrated that this method can extract target information of cells effectively. The U-Net++ method can be further used to analyze the movement of cells and help understanding the biofilm formation in micro-droplets.

Automated summary

By using the U-Net++ neural network model, the image segmentation of bacteria is improved, leading to higher accuracy and robustness in the formation of bacterial biofilms, even in cases of low contrast and noise. This method has great significance for understanding the growth of bacterial biofilms in extreme environments.