Attention-Driven UNet Enhancement for Accurate Segmentation of Growing Bacteria in Microscopy Images

Analyzing microscopy images of large growing cell samples using traditional methods is a complex and time-consuming process. In this work, we have developed an attention-driven UNet-enhanced model using deep learning techniques to efficiently quantify the position, area, and circularity of bacterial spores and vegetative cells from images containing more than 10,000 bacterial cells. Additionally, we present a workflow that includes automated large field-of-view acquisition and semi-automatic stitching of images into a large composite image. Our attention-driven UNet algorithm has an accuracy of 95%, precision of 84%, sensitivity of 73%, and specificity of 98%. Therefore, it can segment cells at a level comparable to manual annotation. We demonstrate the efficacy of this model by applying it to a live-dead decontamination assay. The model is provided in three formats: Python code, a Binder that operates within a web browser without needing installation, and a Flask Web application for local use.