CoChat
Sign Up
Cited by 22

The DOI matches its Crossref record; 3 of the 3 references it cites were found in a registry or an index.

BCM3D 2.0: accurate segmentation of single bacterial cells in dense biofilms using computationally generated intermediate image representations

Zhang, Ji, Wang, Yibo, Donarski, Eric D., Toma, Tanjin T., Miles, Madeline T., Acton, Scott T., Gahlmann, Andreas·2022·npj Biofilms and Microbiomes
DOI 10.1038/s41522-022-00362-4PMC PMC9760640

Published version of a preprint

Open this paper in CoChat — free account

Abstract

Accurate detection and segmentation of single cells in three-dimensional (3D) fluorescence time-lapse images is essential for observing individual cell behaviors in large bacterial communities called biofilms. Recent progress in machine-learning-based image analysis is providing this capability with ever-increasing accuracy. Leveraging the capabilities of deep convolutional neural networks (CNNs), we recently developed bacterial cell morphometry in 3D ( BCM3D ), an integrated image analysis pipeline that combines deep learning with conventional image analysis to detect and segment single biofilm-dwelling cells in 3D fluorescence images. While the first release of BCM3D ( BCM3D 1.0 ) achieved state-of-the-art 3D bacterial cell segmentation accuracies, low signal-to-background ratios (SBRs) and images of very dense biofilms remained challenging. Here, we present BCM3D 2.0 to address this challenge. BCM3D 2.0 is entirely complementary to the approach utilized in BCM3D 1.0 . Instead of training CNNs to perform voxel classification, we trained CNNs to translate 3D fluorescence images into intermediate 3D image representations that are, when combined appropriately, more amenable to conventional mathematical image processing than a single experimental image. Using this approach, improved segmentation results are obtained even for very low SBRs and/or high cell density biofilm images. The improved cell segmentation accuracies in turn enable improved accuracies of tracking individual cells through 3D space and time. This capability opens the door to investigating time-dependent phenomena in bacterial biofilms at the cellular level.

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s41522-022-00362-4

Cites

3 of 3 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Performance of BCM3D 2.0 on previously unseen simulated biofilm images. a Cell counting accuracy (using an Intersection-over-Union (IoU) matching threshold of 0.5 for each segmented object) averaged over N = 10 replicate datasets for cells labeled with cytosolic fluorophores. IoU is a well-documented metric23,30,60 quantifying the amount of overlap between predicted cell and actual cell volumes. b Example image of cells labeled with cytosolic fluorophores (Cell density = 62.2%, SBR = 1.34, indicated by white rectangle in a. a Cell counting accuracy (using an IoU matching threshold of 0.5 for each segmented object) averaged over N = 10 replicate datasets for cells labeled with membrane-localized fluorophores. d Example image of cells labeled with membrane-localized fluorophores (Cell density = 62.2%, SBR = 1.34, indicated by white rectangles in c). The red arrow indicates a close cell-to-cell contact. e, f Comparison of segmentation accuracies achieved by BCM3D 1.0 (CNN + LCuts), Omnipose trained from scratch, Cellpose 2 fine-tuned and BCM3D 2.0 for cytosolic and membrane labeling, respectively (SBR = 1.34, cell density = 62.2%). Segmentation accuracy is parameterized in terms of cell counting accuracy (y axis) and IoU matching threshold (x axis). Each data point is the average of N = 10 independent biofilm images. Data are presented as mean values ± one standard deviation. Scale bars are 2 µm.
Fig. 1 Performance of BCM3D 2.0 on previously unseen simulated biofilm images. a Cell counting accuracy (using an Intersection-over-Union (IoU) matching threshold of 0.5 for each segmented object) averaged over N = 10 replicate datasets for cells labeled with cytosolic fluorophores. IoU is a well-documented metric23,30,60 quantifying the amount of overlap between predicted cell and actual cell volumes. b Example image of cells labeled with cytosolic fluorophores (Cell density = 62.2%, SBR = 1.34, indicated by white rectangle in a. a Cell counting accuracy (using an IoU matching threshold of 0.5 for each segmented object) averaged over N = 10 replicate datasets for cells labeled with membrane-localized fluorophores. d Example image of cells labeled with membrane-localized fluorophores (Cell density = 62.2%, SBR = 1.34, indicated by white rectangles in c). The red arrow indicates a close cell-to-cell contact. e, f Comparison of segmentation accuracies achieved by BCM3D 1.0 (CNN + LCuts), Omnipose trained from scratch, Cellpose 2 fine-tuned and BCM3D 2.0 for cytosolic and membrane labeling, respectively (SBR = 1.34, cell density = 62.2%). Segmentation accuracy is parameterized in terms of cell counting accuracy (y axis) and IoU matching threshold (x axis). Each data point is the average of N = 10 independent biofilm images. Data are presented as mean values ± one standard deviation. Scale bars are 2 µm.CC BY · PubMed Central
Fig. 2 Performance of BCM3D 2.0 on experimental biofilm images. a Principal component analysis of the segmentation objects (obtained from experimentally acquired images) that were classified by a pre-trained 3D CNN as either physiologically reasonable rod-shaped cells or oddly shaped (not-rod-shaped) cells. Three examples cell shapes of each class are shown to the right and left, respectively. Inset: same analysis on simulated biofilm images. Scale bars are 0.6 µm. b 3D segmentation results for objects classified as physiologically reasonable rod-shaped cells. Scale bar is 5 µm. c 3D segmentation results for objects classified as oddly shaped. Scale bar is 5 µm. d, e Comparison of the solidity and minor axis ratio distributions of rod-shaped and oddly shaped objects.
Fig. 2 Performance of BCM3D 2.0 on experimental biofilm images. a Principal component analysis of the segmentation objects (obtained from experimentally acquired images) that were classified by a pre-trained 3D CNN as either physiologically reasonable rod-shaped cells or oddly shaped (not-rod-shaped) cells. Three examples cell shapes of each class are shown to the right and left, respectively. Inset: same analysis on simulated biofilm images. Scale bars are 0.6 µm. b 3D segmentation results for objects classified as physiologically reasonable rod-shaped cells. Scale bar is 5 µm. c 3D segmentation results for objects classified as oddly shaped. Scale bar is 5 µm. d, e Comparison of the solidity and minor axis ratio distributions of rod-shaped and oddly shaped objects.CC BY · PubMed Central
Fig. 3 Multi-cell tracking in simulated biofilms. a Simulated fluorescence time-lapse images of growing E. coli-like biofilm. The SBRs of these images are ~1.65. Contours are color-coded based on segmentation and tracking results. Scale bars are 2 µm. b An example of 3D tracking and lineage tracing for simulated biofilm images. For clarity, spatial trajectories and lineages originating from only a single ancestor cell are displayed. The estimated graph is shown in blue and the corresponding ground truth graph is shown in red. The entire biofilm contains over sixty graphs of this type. c Two AOGM metrics calculated as TRA_edge and TRA_full are plotted against image SBR. The gray dashed line indicates tracking of the GT segmentation using the same tracking algorithm. d Same data as in c plotted as a function of cell counting accuracy at IoU = 0.5, a segmentation accuracy metric that increases for increasing SBR in the raw images23. e Doubling cycle distribution of simulated data and corresponding tracking results. A completed cell cycle is defined as a track in which the parent cell is able to split twice. This threshold results in a lower count numbers of estimated cell division, but does not alter the shape of the distribution. f TRA_edge (left axis) and edge-correction (EC) percentage (right axis) for different temporal sampling rates. EC percentage indicates how many parent-daughter relationships are misassigned based on the tracking results.
Fig. 3 Multi-cell tracking in simulated biofilms. a Simulated fluorescence time-lapse images of growing E. coli-like biofilm. The SBRs of these images are ~1.65. Contours are color-coded based on segmentation and tracking results. Scale bars are 2 µm. b An example of 3D tracking and lineage tracing for simulated biofilm images. For clarity, spatial trajectories and lineages originating from only a single ancestor cell are displayed. The estimated graph is shown in blue and the corresponding ground truth graph is shown in red. The entire biofilm contains over sixty graphs of this type. c Two AOGM metrics calculated as TRA_edge and TRA_full are plotted against image SBR. The gray dashed line indicates tracking of the GT segmentation using the same tracking algorithm. d Same data as in c plotted as a function of cell counting accuracy at IoU = 0.5, a segmentation accuracy metric that increases for increasing SBR in the raw images23. e Doubling cycle distribution of simulated data and corresponding tracking results. A completed cell cycle is defined as a track in which the parent cell is able to split twice. This threshold results in a lower count numbers of estimated cell division, but does not alter the shape of the distribution. f TRA_edge (left axis) and edge-correction (EC) percentage (right axis) for different temporal sampling rates. EC percentage indicates how many parent-daughter relationships are misassigned based on the tracking results.CC BY · PubMed Central
Fig. 4 Multi-cell tracking of experimentally acquired S. oneidensis biofilm images. a Experimentally acquired fluorescence time-lapse images of a growing S. oneidensis biofilm with overlaid single-cell segmentation contours. Images were acquired every 30 s for 15 min. Corresponding cells in different frames are displayed in the same color. Scale bar is 2 µm. b Individual cell trajectories in the biofilm shown in a. Cells move very little during the short 15-min imaging time. Scale bar is 1 µm. c Cell volumes over time for four example cells. Cell division and dispersion events are indicated for each trajectory. Single-cell growth rate was measured by calculating the slope of linearly fitted line for each curve. d Experimentally acquired fluorescence time-lapse images of a growing S. oneidensis biofilm with overlaid single-cell segmentation contours. Images were acquired every five minutes for five hours. Corresponding cells in different frames are displayed in the same color. Scale bar is 2 µm. e Individual cell trajectories in the biofilm shown in a. Cell displacements are more pronounced over the 5-h imaging time. Scale bar is 1 µm. f Evolution of cell volumes over time for a single selected cell. The solid line represents a manually annotated trajectory, and the dashed line represents the nearest neighbor tracking trajectory of the same initial cell. Cell division and dispersion events are indicated on each trajectory. The single-cell growth rate was measured by calculating the slope of the trajectory segment between consecutive cell division events. The red arrow and the boxes indicate periodic underestimation of the cell volume due to oversegmentation.
Fig. 4 Multi-cell tracking of experimentally acquired S. oneidensis biofilm images. a Experimentally acquired fluorescence time-lapse images of a growing S. oneidensis biofilm with overlaid single-cell segmentation contours. Images were acquired every 30 s for 15 min. Corresponding cells in different frames are displayed in the same color. Scale bar is 2 µm. b Individual cell trajectories in the biofilm shown in a. Cells move very little during the short 15-min imaging time. Scale bar is 1 µm. c Cell volumes over time for four example cells. Cell division and dispersion events are indicated for each trajectory. Single-cell growth rate was measured by calculating the slope of linearly fitted line for each curve. d Experimentally acquired fluorescence time-lapse images of a growing S. oneidensis biofilm with overlaid single-cell segmentation contours. Images were acquired every five minutes for five hours. Corresponding cells in different frames are displayed in the same color. Scale bar is 2 µm. e Individual cell trajectories in the biofilm shown in a. Cell displacements are more pronounced over the 5-h imaging time. Scale bar is 1 µm. f Evolution of cell volumes over time for a single selected cell. The solid line represents a manually annotated trajectory, and the dashed line represents the nearest neighbor tracking trajectory of the same initial cell. Cell division and dispersion events are indicated on each trajectory. The single-cell growth rate was measured by calculating the slope of the trajectory segment between consecutive cell division events. The red arrow and the boxes indicate periodic underestimation of the cell volume due to oversegmentation.CC BY · PubMed Central
Fig. 5 Simulation of fluorescent biofilms images and intermediate image representations. a Cell arrangements obtained by CellModeller. b Simulated 3D fluorescence image based on the cell arrangements in a. c Ground truth information of a 2D slice. Different cells are shown in different colors and intercellular spaces (background voxels) are displayed in black. d 2D slices of the simulated fluorescence image corresponding to the ground truth shown in c. The upper panel shows cells containing cytosolic fluorophores, the lower panel shows cells with fluorescently stained membranes. e, f Intermediate image representations generated from the ground truth information shown in c. See text for details. Scale bars are 5 µm.
Fig. 5 Simulation of fluorescent biofilms images and intermediate image representations. a Cell arrangements obtained by CellModeller. b Simulated 3D fluorescence image based on the cell arrangements in a. c Ground truth information of a 2D slice. Different cells are shown in different colors and intercellular spaces (background voxels) are displayed in black. d 2D slices of the simulated fluorescence image corresponding to the ground truth shown in c. The upper panel shows cells containing cytosolic fluorophores, the lower panel shows cells with fluorescently stained membranes. e, f Intermediate image representations generated from the ground truth information shown in c. See text for details. Scale bars are 5 µm.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 3 references

Research works printed but not linked: 55

  1. Nadell, C. D., Drescher, K. & Foster, K. R. Spatial structure, cooperation and competition in biofilms. Nat. Rev. Microbiol. 14, 589 (2016). · DOI
  2. Flemming, H. C. et al. Biofilms: an emergent form of bacterial life. Nat. Rev. Microbiol 14, 563–575 (2016). · DOI · Paper page
  3. Jayathilake, P. G. et al. A mechanistic individual-based model of microbial communities. PLoS One 12, e0181965 (2017). · DOI
  4. Persat, A. et al. The mechanical world of bacteria. Cell 161, 988–997 (2015). · DOI
  5. De la Fuente-Núñez, C., Reffuveille, F., Fernández, L. & Hancock, R. E. Bacterial biofilm development as a multicellular adaptation: antibiotic resistance and new therapeutic strategies. Curr. Opin. Microbiol. 16, 580–589 (2013). · DOI
  6. Frost, I. et al. Cooperation, competition and antibiotic resistance in bacterial colonies. ISME J. 12, 1582–1593 (2018). · DOI
  7. Yan, J. & Bassler, B. L. Surviving as a community: antibiotic tolerance and persistence in bacterial biofilms. Cell Host Microbe 26, 15–21 (2019). · DOI
  8. Stelzer, E. H. Light-sheet fluorescence microscopy for quantitative biology. Nat. Methods 12, 23–26 (2015). · DOI
  9. Power, R. M. & Huisken, J. A guide to light-sheet fluorescence microscopy for multiscale imaging. Nat. Methods 14, 360–373 (2017). · DOI
  10. Voleti, V. et al. Real-time volumetric microscopy of in vivo dynamics and large-scale samples with SCAPE 2.0. Nat. Methods 16, 1054–1062 (2019). · DOI · Paper page
  11. Mimori-Kiyosue, Y. Imaging mitotic processes in three dimensions with lattice light-sheet microscopy. Chromosome Res. 29, 37–50 (2021). · DOI
  12. Rosenberg, J., Cao, G., Borja-Prieto, F. & Huang, J. Lattice light-sheet microscopy multi-dimensional analyses (LaMDA) of T-cell receptor dynamics predict T-cell signaling states. Cell Syst. 10, 433–444.e435 (2020). · DOI
  13. Manley, H. R. et al. Frontline Science: dynamic cellular and subcellular features of migrating leukocytes revealed by in vivo lattice lightsheet microscopy. J. Leukoc. Biol. 108, 455–468 (2020). · DOI
  14. Yordanov, S. et al. Single-objective high-resolution confocal light sheet fluorescence microscopy for standard biological sample geometries. Biomed. Opt. Express 12, 3372–3391 (2021). · DOI
  15. Qin, B., et al. Cell position fates and collective fountain flow in bacterial biofilms revealed by light-sheet microscopy. Science 369, eabb8501 (2020). · DOI
  16. Parthasarathy, R. Monitoring microbial communities using light sheet fluorescence microscopy. Curr. Opin. Microbiol. 43, 31–37 (2018). · DOI
  17. Heintzmann, R. & Huser, T. Super-resolution structured illumination microscopy. Chem. Rev. 117, 13890–13908 (2017). · DOI
  18. Gustafsson, M. G. et al. Three-dimensional resolution doubling in wide-field fluorescence microscopy by structured illumination. Biophys. J. 94, 4957–4970 (2008). · DOI
  19. Weigert, M. et al. Content-aware image restoration: pushing the limits of fluorescence microscopy. Nat. Methods 15, 1090–1097 (2018). · DOI
  20. Chen, J. et al. Three-dimensional residual channel attention networks denoise and sharpen fluorescence microscopy image volumes. Nat. Methods 18, 678–687 (2021). · DOI
  21. Moen, E. et al. Deep learning for cellular image analysis. Nat Methods 16, 1233–1246 (2019). · DOI · Paper page
  22. Jeckel, H. & Drescher, K. Advances and opportunities in image analysis of bacterial cells and communities. FEMS Microbiol. Rev. 45, fuaa062 (2020). · DOI
  23. Zhang, M. et al. Non-invasive single-cell morphometry in living bacterial biofilms. Nat. Commun. 11, 6151 (2020). · DOI
  24. Wang, J. et al. LCuts: linear clustering of bacteria using recursive graph cuts. in 2019 IEEE International Conference on Image Processing (ICIP) 1575–1579 (IEEE, 2019). · DOI
  25. Stringer, C., Wang, T., Michaelos, M. & Pachitariu, M. Cellpose: a generalist algorithm for cellular segmentation. Nat. Methods 18, 100–106 (2021). · DOI · Paper page
  26. Weigert, M. et al. Star-convex polyhedra for 3d object detection and segmentation in microscopy. in Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision https://arxiv.org/abs/1908.03636 (2020). · DOI
  27. Scherr, T., Löffler, K., Böhland, M. & Mikut, R. Cell segmentation and tracking using CNN-based distance predictions and a graph-based matching strategy. Plos One 15, e0243219 (2020). · DOI
  28. Meyer, F. & Beucher, S. Morphological segmentation. J. Vis. Commun. Image Represent. 1, 21–46 (1990). · DOI
  29. Otsu, N. A threshold selection method from gray-level histograms. IEEE Trans. Syst. Man, Cybern. 9, 62–66 (1979). · DOI
  30. Weigert, M. et al. Star-convex polyhedra for 3D object detection and segmentation in microscopy. in 2020 IEEE Winter Conference on Applications of Computer Vision (WACV) 3655–3662 (2020). · DOI
  31. Laasmaa, M., Vendelin, M. & Peterson, P. Application of regularized Richardson–Lucy algorithm for deconvolution of confocal microscopy images. J. Microsc. 243, 124–140 (2011). · DOI
  32. Zhang, J. et al. Optically accessible microfluidic flow channels for noninvasive high-resolution biofilm imaging using lattice light sheet microscopy. J. Phys. Chem. B 125, 12187–12196 (2021). · DOI
  33. Liu, R. & Jia, J. Reducing boundary artifacts in image deconvolution. in 2008 15th IEEE International Conference on Image Processing 505–508 (IEEE, 2008). · DOI
  34. Starck, J.-L., Pantin, E. & Murtagh, F. Deconvolution in astronomy: a review. Publ. Astron. Soc. Pac. 114, 1051 (2002). · DOI
  35. Wang, J. et al. Graph-theoretic post-processing of segmentation with application to dense biofilms. IEEE Trans. Image Process. 30, 8580–8594 (2021). · DOI
  36. Cutler, K. J., et al. Omnipose: a high-precision morphology-independent solution for bacterial cell segmentation. Nat. Methods 19, 1438–1448 (2022). · DOI
  37. Matula, P. et al. Cell tracking accuracy measurement based on comparison of acyclic oriented graphs. PLoS One 10, e0144959 (2015). · DOI
  38. Ulman, V. et al. An objective comparison of cell-tracking algorithms. Nat. Methods 14, 1141–1152 (2017). · DOI
  39. Wen, C. et al. 3DeeCellTracker, a deep learning-based pipeline for segmenting and tracking cells in 3D time lapse images. Elife 10, e59187 (2021). · DOI
  40. Magnusson, K. E., Jalden, J., Gilbert, P. M. & Blau, H. M. Global linking of cell tracks using the Viterbi algorithm. IEEE Trans. Med. Imaging 34, 911–929 (2015). · DOI
  41. Moen, E. et al. Accurate cell tracking and lineage construction in live-cell imaging experiments with deep learning. Preprint at https://www.biorxiv.org/content/10.1101/803205v2 (2019). · DOI
  42. Thormann, K. M. et al. Initial Phases of biofilm formation in Shewanella oneidensis MR-1. J. Bacteriol. 186, 8096–8104 (2004). · DOI
  43. Gao, R. et al. Cortical column and whole-brain imaging with molecular contrast and nanoscale resolution. Science 363, eaau8302 (2019). · DOI
  44. Schindelin, J. et al. Fiji: an open-source platform for biological-image analysis. Nat. Methods 9, 676–682 (2012). · DOI · Paper page
  45. Pettersen, E. F. et al. UCSF ChimeraX: structure visualization for researchers, educators, and developers. Protein Sci. 30, 70–82 (2021). · DOI
  46. Parslow, A., Cardona, A. & Bryson-Richardson, R.J. Sample drift correction following 4D confocal time-lapse imaging. J. Vis. Exp. 51086 (2014). · DOI
  47. Rudge, T. J., Steiner, P. J., Phillips, A. & Haseloff, J. Computational modeling of synthetic microbial biofilms. ACS Synth. Biol. 1, 345–352 (2012). · DOI
  48. Venkateswaran, K. et al. Polyphasic taxonomy of the genus Shewanella and description of Shewanella oneidensis sp. nov. Int. J. Syst. Bacteriol. 49, 705–724 (1999). · DOI
  49. Yan, T., Richardson, C. J., Zhang, M. & Gahlmann, A. Computational correction of spatially variant optical aberrations in 3D single-molecule localization microscopy. Opt. Express 27, 12582–12599 (2019). · DOI
  50. Liao, P.-S., Chen, T.-S. & Chung, P.-C. A fast algorithm for multilevel thresholding. J. Inf. Sci. Eng. 17, 713–727 (2001).
  51. Van der Walt, S. et al. scikit-image: image processing in Python. PeerJ 2, e453 (2014). · DOI
  52. Zunair, H., Rahman, A., Mohammed, N. & Cohen, J. P. Uniformizing techniques to process CT scans with 3D CNNs for tuberculosis prediction. in International Workshop on PRedictive Intelligence In MEdicine 156–168 (Springer, 2020). · DOI
  53. Jaccard, P. The distribution of the flora in the alpine zone. 1. N. Phytol. 11, 37–50 (1912). · DOI · Paper page
  54. Csurka, G., Larlus, D., Perronnin, F. & Meylan, F. What is a good evaluation measure for semantic segmentation? BMVC 27, 10.5244 (2013).
  55. Laine, R. F., Arganda-Carreras, I., Henriques, R. & Jacquemet, G. Avoiding a replication crisis in deep-learning-based bioimage analysis. Nat. Methods 18, 1136–1144 (2021). · DOI

Cited by

Sorted by citations

Metadata

Authors
Zhang, J., Wang, Y., Donarski, E. D., Toma, T. T., Miles, M. T., Acton, S. T., & Gahlmann, A.
Venue
npj Biofilms and Microbiomes
Publisher
Springer Science and Business Media LLC
Published
2022-12-18
Volume
8
Issue
1
ISSN
2055-5008
DOI
10.1038/s41522-022-00362-4
PMC
PMC9760640
Cited by
22
References
3
Language
en
Licence
Funders
U.S. Department of Health & Human Services | NIH | National Institute of General Medical Sciences (1R01GM139002)

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review