SyMBac: Synthetic Micrographs for Training Deep Neural Networks for Accurate Segmentation of Bacterial Cells Microscope model for generating phase-contrast and fluorescence

We present a novel method of bacterial image segmentation using machine learning based on Synthetic Micrographs of Bacteria (SyMBac). SyMBac allows for rapid, automatic creation of arbitrary amounts of training data that combines detailed models of cell growth, physical interactions, and microscope optics to create synthetic images which closely resemble real micrographs, with access to the ground truth positions of cells. This approach provides the following major advantages: (1) Synthetic training data can be generated virtually instantly, and on demand; (2) it is accompanied by perfect ground truth information, which also means large sets of training data can be generated without the need for curation; (3) Biological ground truth and imaging effects can be tuned independently, enabling easy adaptation to different biological conditions, imaging platforms, and imaging modalities. Machine-learning models trained on SyMBac data generate more accurate and precise cell masks than those trained on human annotated data. This enables precise analysis of cell size regulation along changing conditions in a growing bacterial cell culture, revealing novel insights about the physiological dynamics of individual bacterial cells during entry and exit from dormancy. by perfect ground truth information, which means large sets of training data can be generated without the need for curation. Our benchmarking results show that deep-learning algorithms trained with SyMBac are far superior, in terms of accuracy and precision of segmentation, compared to the same models trained on human annotated or computationally segmented data. This will significantly advance the field of quantitative microbiology.