Omnipose: a high-precision morphology-independent solution for bacterial cell segmentation.
Cutler KJ, Stringer C, Lo TW, Rappez L, Stroustrup N, Brook Peterson S, Wiggins PA, Mougous JD. · 2022
Read Omnipose: a high-precision morphology-independent solution for bacterial cell segmentation. by Cutler KJ, Stringer C, Lo TW, Rappez L, Stroustrup N, Brook Peterson S, Wiggins PA, Mougous JD. (2022) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc9636021.txt.
Introduction
Cutler and co-authors present Omnipose, a deep network tool for segmenting bacterial cells in images. Microscopy can now track form and molecules in single bacteria. Faithful cell outlines are the step that gates all such work. Odd cell shapes and optics break older outline tools.
Omnipose uses gradient outputs from a distance field to trace borders. That design fixes errors its forerunner Cellpose made on hard cells. Tests covered mixed bacterial cultures side by side. Antibiotic treated cells with warped shapes were segmented well. Long and branched cell forms also came through clean. Gains reached non-bacterial subjects and varied image types. Three-dimensional objects were handled by the same tool. A final test read shape changes in cells under attack by rival bacteria. The authors bill it as a shape proof route to cell outlines.
Catalog record
| CREATOR | Cutler KJ, Stringer C, Lo TW, Rappez L, Stroustrup N, Brook Peterson S, Wiggins PA, Mougous JD. |
|---|---|
| YEAR | 2022 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 90 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Omnipose: a high-precision morphology-independent solution for bacterial cell segmentation By Kevin J Cutler, Carsen Stringer, Teresa W Lo, Luca Rappez, Nicholas Stroustrup, S Brook Peterson, Paul A Wiggins, Joseph D Mougous Published in: Nature Methods DOI: 10.1038/s41592-022-01639-4 ABSTRACT Abstract Advances in microscopy hold great promise for allowing quantitative and precise measurement of morphological and molecular phenomena at the single-cell level in bacteria; however, the potential of this approach is ultimately limited by the availability of methods to faithfully segment cells independent of their morphological or optical characteristics. Here, we present Omnipose, a deep neural network image-segmentation algorithm. Unique network outputs such as the gradient of the distance field allow Omnipose to accurately segment cells on which current algorithms, including its predecessor, Cellpose, produce errors. We show that Omnipose achieves unprecedented segmentation performance on mixed bacterial cultures, antibiotic-treated cells and cells of elongated or branched morphology. Furthermore, the benefits of Omnipose extend to non-bacterial subjects, varied imaging modalities and three-dimensional objects. Finally, we demonstrate the utility of Omnipose in the characterization of extreme morphological phenotypes that arise during interbacterial antagonism. Our results distinguish Omnipose as a powerful tool for characterizing diverse and arbitrarily shaped cell types from imaging data. Subject terms: Bacteria, Imaging MAIN Although light microscopy is a valuable tool for characterizing cellular and subcellular structures and dynamics, quantitative analysis of microscopy images remains a persistent challenge1. This is especially pertinent to the study of bacteria, many of which have dimensions in the range of visible wavelengths. Thus, their cell body is composed of a small number of pixels (for example, ~100–300 px2 for Escherichia coli in typical experiments). At this scale, accurate subcellular localization requires defining the cell boundary with single-pixel precision. The process of defining boundaries within images is termed segmentation, and this is a critical first step in image analysis pipelines2,3. In addition to their small size, bacteria adopt a wide range of morphologies. Although many commonly studied bacteria are well approximated by rods or spheres, there is growing interest in bacteria with more elaborate shapes4. Some examples include Streptomycetales, which form long filamentous and branched structures5, and Caulobacterales, which possess extended appendages distinct from their cytoplasm6. Furthermore, microfluidic devices are allowing researchers to capture the responses of bacteria to assorted treatments such as antibiotics, which often result in highly irregular morphologies7. Whether native or induced, atypical cell morphologies present a distinct problem at the cell-segmentation phase of image analysis8,9. This is compounded when such cells are present alongside those adopting other morphologies, as is the case in many natural samples of interest10. Although some algorithms can be trained to segment objects imaged using assorted modalities, to date there is no generalizable solution for segmenting bacterial cells of assorted size and shape1. Cell segmentation is a complex problem that extends beyond microbiological research; thus, many solutions are currently available in image-analysis programs8,9,11–27. Most of these solutions use traditional image-processing techniques such as intensity thresholding to segment isolated cells; however, this approach performs poorly on cells in close contact and it requires image-by-image tuning to optimize parameters. SuperSegger was developed to address thresholding issues specifically in bacterial phase-contrast images13. This program utilizes both traditional image filtering techniques and a shallow neural network to correct for errors that thresholding and watershed segmentation tend to produce. Deep neural networks (DNNs) are now widely recognized as superior tools for cell segmentation28. Unlike traditional image processing, machine-learning approaches such as DNNs require training on a ground-truth dataset of cells and corresponding labels. Trained DNNs are thus limited in applicability to images that are representative of those in the training dataset. Early DNN approaches were based on the Mask R-CNN architecture24, whereas more recent algorithms such as StarDist, Cellpose and MiSiC are based on the U-net architecture12,15,26. Pachitariu and colleagues showed that Cellpose outperforms Mask R-CNN and StarDist on a variety of cell types and cell-like objects, distinguishing it as a general solution for cell segmentation12. Notably, the representation of bacteria in their study was limited. MiSiC was developed as a general DNN-based solution for bacterial segmentation; however, the authors of MiSiC did not provide comparisons to other DNN algorithms15. Here, we evaluated the performance of state-of-the-art cell segmentation algorithms on a diverse collection of bacterial cells. Our findings motivated the design of a new algorithm, Omnipose, which substantially outperforms all segmentation algorithms tested across a wide range of bacterial cell sizes, morphologies, and optical characteristics. We have made Omnipose and all associated data available to researchers and we anticipate that our model, without retraining, can be applied to diverse bacteriological systems. Furthermore, following the incorporation of additional ground-truth data, Omnipose could serve as a platform for segmenting various eukaryotic cells and extended, anisotropic objects more broadly. RESULTS Evaluation of bacterial cell segmentation algorithms Numerous image segmentation algorithms have been developed and the performance of many of these on bacterial cells is documented1. These broadly fall into three categories: (1) traditional image-processing approaches (for example, thresholding and watershed); (2) traditional/machine-learning hybrid approaches; and (3) DNN approaches. Given the goal of developing software with the capacity to recognize bacteria universally, we sought to identify strongly performing algorithms for further development. An unbiased, quantitative comparison of cell segmentation algorithms on bacterial cells has not been performed; thus, we selected one or more representatives from each category for our analysis: Morphometrics23 (1); SuperSegger13 (2); and Mask R-CNN27, StarDist26, MiSiC15, and Cellpose12 (3). A detailed review of these choices is in the Methods. For training and benchmarking of these algorithms, we acquired micrographs of assorted bacterial species representing diverse morphologies and optical characteristics. Many studies of bacteria involve mutations or treatments that cause extreme morphologies. To capture this additional diversity, we included wild-type and mutant bacteria grown in the presence of two β-lactam antibiotics, cephalexin and aztreonam, and A22, which targets MreB29. Finally, based on our interest in microbial communities, we acquired images of mixtures of bacterial species which display distinct morphologies and optical characteristics. In total, we collected 4,833 images constituting approximately 700,900 individual cells deriving from 14 species (Supplementary Table 1). Next, we developed a streamlined approach for manual cell annotation and applied it to these images (Methods), yielding 47,000 representative annotated cells that serve as our ground-truth dataset (bact_phase). We divided this data into a 27,500-cell training set and a 19,500-cell benchmarking set. Relevant cellular metrics (area, perimeter and mean diameter) did not differ substantially between the groups, confirming that the benchmarking set faithfully represents the training set (Extended Data Fig. 1). [Figure: Size and morphology metrics are indistinguishable between cell populations included in training and test datasets.(a) Mean cell diameter, (b) cell area, and (c) cell perimeter calculated for our bacterial phase-contrast ground-truth dataset. P-values are displayed for the two-sided KS test. n = 47,000 (27,500 for training, 19,500 for testing). (d). Comparison of diameter metrics of a timelapse of elongated cell growth. The Cellpose diameter metric is the diameter of a circle with equivalent area. Omnipose diameter metric is proportional to the mean of the distance transform (see Methods). (e) Bacteria displayed are a single population analyzed of A. baylyi transformed with a ΔftsN::kan PCR fragment. Yellow lines indicate cell label boundaries. Scale bar is 1 μm.] To facilitate direct comparison of