Bacterial Counting: Quick, easy and accurate?
Download as PPSX, PDF1 like461 views
The document discusses the methods for bacterial counting, comparing traditional manual techniques with automated image analysis approaches utilizing software like Image Pro Plus. It highlights results showing that automated counting significantly reduces time without sacrificing accuracy, while also addressing limitations of current freeware and software. Additionally, it covers various applications and implications for conservation efforts and ecological research.
Bacterial Counting: Quick, easy and accurate?
- 1. Bacterial Counting: Quick, easy and accurate? Kunnen, T. H. Moodley, G. K. Robertson-Andersson, D. V. University of KwaZulu-Natal, School of Life Science
- 2. Overview Introduction • Microbial loop • Bacterial numbers and biomass • Image analysis • Freeware • Macro coding Conclusions • Advantages vs. Disadvantages • Conservation efforts • Other applications Materials and Methods • Macro coding • Repeated automated counting and sizing • Binary Segmentation • Testing the system Results and Discussion • Human data vs. Automatic data • Time differences
- 3. Introduction • The Microbial Loop • Cyclic interaction • Trophic linkages • Bacterial numbers, biomass and productivity • Nutrient cycling Figure 1: The microbial loop as conceptualized by Landry and Kirchman (2002)
- 4. Introduction contd… Figure 2: Adapted simplified marine trophic pyramid (www1)
- 5. Introduction contd… Figure 2: Adapted simplified marine trophic pyramid (www1)
- 6. Introduction contd... • Traditional bacterial enumeration • Photo enlargement • Nucleic stains, PC’s • Image analysis software • Many freeware options • Recognition errors in counting cells > 0,75 µm • 53 % www2 www3
- 7. Introduction contd... • CellProfiler (Carpenter et al., 2006; + 3340 more journal articles) • Headed by Anne Carpenter • Whitehead Institute, USA • Pipelines • CellC (Selinummi et al., 2005; + 79 more journal articles) • Written by Jyriki Selinummi for ISB Seattle • Calibration • Wählby Lab (Sadanandan et al., 2016) • Headed by Carolina Wählby • Uppsala University, Sweden • 2013 + 2015 • “I'm not sure how well I calibrated the analysis for size”
- 8. Introduction contd... • Limitations of freeware • On point functionality • Outdated software and hardware • Limited or no technical support • Website closed down / domain inaccessible • Author (s) / programmer no longer available www4 www5 www6
- 9. Introduction contd... • Image analysis based on Binary Segmentation • Automated image analysis by binary segmentation (Krambeck et al., 1981) • Commercially available image analysis software – Image Pro Plus (IPP) • Macro scripting does what it is told and has the potential to save time and reduce human bias
- 10. Materials and Methods • Coding for automated Z- stacking of unfocused images using IPP EDF (Extended Depth of Field) This is 48 lines of code of the 2145 = 2%
- 11. Materials and Methods contd... • Coding for repeated counting and sizing of bacteria within existing commercially available image analysis software IPP This is 51 lines of code of the 242 = 21%
- 12. IPP Repeated Automatic Counting
- 13. Materials and Methods contd... Binary segmentation with histogram selection Background noise Data
- 15. Materials and Methods contd... • 8 volunteers given basic training on IPP • 60 repeated random bacterial images were supplied to each volunteer • Volunteers required to time themselves while counting and sizing (length and width) “objects” they classify as being bacterial cells • Mandatory 2/3 day break • 10 repeated random bacterial images extracted from the 60 and volunteers required to time themselves while they re-count and size • Directly after, volunteers used the IPP macro to automatically count and size “objects” within threshold limits (27-87) using increments of 10 Testing the system
- 16. Results and Discussion • No difference between human vs. automated analysis for numbers and biomass overall • Mean time reduction of total time (1136.83 %) and time per cell (822.25 %) of for 8 volunteers for automated analysis • Equates to average total time differences of 5.06 hr manual vs. 26.71 min auto • Real time of 2 days vs. 2 hr
- 17. Results and Discussion • Colour blindness – Surprising outcome! • One volunteer was color blind • Significant impact on segmentation selection • 86 % www9-11
- 18. Conclusions Disadvantages Advantages Manual Automated Slow Become narrow minded when counting Some images require editing Become stricter Non-specific Miss cells entirely Accurate counts vs. accurate biomass Am I counting individual cells? Are individual cells being counted? Manual Automated You know what you counted Fast and relatively easy Judge individual cells accordingly Reproducible It counts and sizes what you tell it to Non-specific Reduces the influence of the halo effect
- 20. Conservation Efforts Wastewater • Principal of the microbial loop • Recycle our water resources www17 www18 Aerobic digestion Dried sludge
- 21. Conservation Efforts Wastewater www12 www13 • Wastewater effluent testing in conjunction with BOD and COD • BOD: up to 20 days to test • COD: less time, requires strong oxidising chemicals
- 23. Conservation Efforts • All macros are currently being applied in the area of microplastic research • Mullet • Sea Urchins • Mussels • Biofilm growth • All macros also being applied to: • Abalone aquaculture (health and safety) • Kosi Bay ecosystem
- 24. Additional Applications • New macros are being attempted to track and trace ragged tooth shark fingerprint markings • Assessment of bacterial loading • Rivers • Estuaries • Oceans • Landfill leachate assessment of bacterial loading • General water quality assessments • Microplastic counting and sizing
- 25. CAN COMPUTERS COUNT BACTERIA? Simpler Better Faster Thank you Acknowledgements Thank you to the MACE lab volunteers and to the NRF for funding this project. Thanks also go to Theo van Zyl, Riaan Rossow, Bertrand Denoix and Kevin Payne.
- 26. References • Carpenter, A. E., Jones, T. R., Lamprecht, M. R., Clarke, C., Kang, I. H., Friman, O., Guertin, D. A., Chang, J. H., Lindquist, R. A., Moffat, J., Golland, P. and Sabatini, D. M. 2006. CellProfiler: image analysis software for identifying and quantifying cell phenotypes. Genome Biol, 7 (10). R100. • Eduard , W., Blomquist, G., Nelson, B. H., Heldal, K. K. 2001. Recognition errors in the quantification of micro– organisms by fluorescence microscopy. Annals of Occupational Hygiene 45: 493–498. • Krambeck, C., Krambeck, H. J. and Overbeck, J. 1981. Microcomputer assisted biomass determination of plankton bacteria on scanning electron micrographs. Applied and Environmental Microbiology, 42. 142-149. • Landry, M. R. and Kirchman, D. L. 2002. Microbial community structure and variability in the tropical Pacific. Deep-Sea Research II, 49. 2669-2693. • Sadanandan, S. K., Baltekin, Ö, Magnusson, K. E. G., Boucharin, A., Ranefall, P., Jaldén, J., Elf, J. and Wählby, C. 2016. IEEE Journal of Selected Topics in Signal Processing, 10 (1). 174-184, • Selinummi, J., Seppälä, J., Yli-Harja, O. and Puhakka, J. 2005. Software for quantification of labeled bacteria from digital microscope images by automated image analysis. BioTechniques, 39. 859-863. Full list of internet images and GIF’s available upon request
Editor's Notes
- #3: The microbial loop is as its name suggests an integrated loop within the microbial fraction of aquatic microorganisms. It is the interaction between bacteria, ciliates, flagellates, microzooplankton and zooplankton and the linking to the primary producers. Although all parts have a role to play, the core are the bacteria and their ability to recycle waste organic matter (which includes all essential macro and minor nutrients) into a form that is usable by incorporating it into their biomass. In this way nutrients and essential growth limiting nutrients are not lost from the system, but are recycled up the food web to higher trophic organisms.
- #4: The microbial loop is as its name suggests an integrated loop within the microbial fraction of aquatic microorganisms. It is the interaction between bacteria, ciliates, flagellates, microzooplankton and zooplankton and the linking to the primary producers. Although all parts have a role to play, the core are the bacteria and their ability to recycle waste organic matter (which includes all essential macro and minor nutrients) into a form that is usable by incorporating it into their biomass. In this way nutrients and essential growth limiting nutrients are not lost from the system, but are recycled up the food web to higher trophic organisms.
- #6: The microbial loop is as its name suggests an integrated loop within the microbial fraction of aquatic microorganisms. It is the interaction between bacteria, ciliates, flagellates, microzooplankton and zooplankton and the linking to the primary producers. Although all parts have a role to play, the core are the bacteria and their ability to recycle waste organic matter (which includes all essential macro and minor nutrients) into a form that is usable by incorporating it into their biomass. In this way nutrients and essential growth limiting nutrients are not lost from the system, but are recycled up the food web to higher trophic organisms.
- #7: Krambeck et al. 1981
- #11: I wrote a series of macro scripts for the commercially available image analysis software Image Pro Plus for repeated counting and sizing of Z stacked images Although these macros do automate the counting and sizing, no computer software, except AI, can gauge what is a bacterial cell and define the limits of what to count. But will get to that a bit later
- #12: I wrote a series of macro scripts for the commercially available image analysis software Image Pro Plus for repeated counting and sizing of Z stacked images Although these macros do automate the counting and sizing, no computer software, except AI, can gauge what is a bacterial cell and define the limits of what to count. But will get to that a bit later
- #13: As this example shows, the steps involve taking a Z stacked image, converting this image into greyscale and then the user selects a histogram segmentation level telling the system what must be counted. Now you will notice that a segmentation level of 77 has been selected and all these cells have not been counted The segmentation level needs to be changed based on environmental situations, different sample preparations or even to much background light Irrespective of the reason, the level needs to be changed to accommodate counting and sizing of as many identified cells as possible
- #14: Previously I said that the image needs to be converted into a greyscale and that is because the system works based on binary segmentation and the user input comes from the histogram selection of that binary segmentation. The big peak is all the background noise and everything on the right is possible data Now this histogram changes with the sample, sometimes the peak will be more to the right, sometimes to the left this is where the user comes in. By changing the histogram selection from 67 to 87, the user has told the system to reduce the amount of background emission to be included in when identifying objects and has therefore become more accurate. To test whether this automatic system of counting and sizing is comparable to human manual counting, we took 7 volunteers, including myself To each volunteer we gave the same randomly selected 60 images, which were randomly mixed for each volunteer They first (after some training) manually counted all objects that they perceived as a bacterial cell Secondly, after a few days break they re-counted the same randomized 10 images from the original pool. And finally they used the macro script to automatically count and size, drawing on their manual sizing skills to determine what should be counted.
- #15: Previously I said that the image needs to be converted into a greyscale and that is because the system works based on binary segmentation and the user input comes from the histogram selection of that binary segmentation. The big peak is all the background noise and everything on the right is possible data Now this histogram changes with the sample, sometimes the peak will be more to the right, sometimes to the left this is where the user comes in. By changing the histogram selection from 67 to 87, the user has told the system to reduce the amount of background emission to be included in when identifying objects and has therefore become more accurate. To test whether this automatic system of counting and sizing is comparable to human manual counting, we took 7 volunteers, including myself To each volunteer we gave the same randomly selected 60 images, which were randomly mixed for each volunteer They first (after some training) manually counted all objects that they perceived as a bacterial cell Secondly, after a few days break they re-counted the same randomized 10 images from the original pool. And finally they used the macro script to automatically count and size, drawing on their manual sizing skills to determine what should be counted.
- #17: What we found was that there was a significant comparable result manual vs automated analysis for both counting and sizing cells, statistically speaking there was no difference between the two Where there was a significant difference was in the time Averaged amongst all 7 volunteers, using the same 60 images, and using automated analysis, was 1155% faster compared to manual analysis! To put this in terms of pure counting time is 4,84 hrs vs 26,92 mins And a real time of 2 days vs 2 hrs
- #18: What we found was that there was a significant comparable result manual vs automated analysis for both counting and sizing cells, statistically speaking there was no difference between the two Where there was a significant difference was in the time Averaged amongst all 7 volunteers, using the same 60 images, and using automated analysis, was 1155% faster compared to manual analysis! To put this in terms of pure counting time is 4,84 hrs vs 26,92 mins And a real time of 2 days vs 2 hrs
- #19: What we found was that there was a significant comparable result manual vs automated analysis for both counting and sizing cells, statistically speaking there was no difference between the two Where there was a significant difference was in the time Averaged amongst all 7 volunteers, using the same 60 images, and using automated analysis, was 1155% faster compared to manual analysis! To put this in terms of pure counting time is 4,84 hrs vs 26,92 mins And a real time of 2 days vs 2 hrs
- #25: What we found was that there was a significant comparable result manual vs automated analysis for both counting and sizing cells, statistically speaking there was no difference between the two Where there was a significant difference was in the time Averaged amongst all 7 volunteers, using the same 60 images, and using automated analysis, was 1155% faster compared to manual analysis! To put this in terms of pure counting time is 4,84 hrs vs 26,92 mins And a real time of 2 days vs 2 hrs
- #27: What we found was that there was a significant comparable result manual vs automated analysis for both counting and sizing cells, statistically speaking there was no difference between the two Where there was a significant difference was in the time Averaged amongst all 7 volunteers, using the same 60 images, and using automated analysis, was 1155% faster compared to manual analysis! To put this in terms of pure counting time is 4,84 hrs vs 26,92 mins And a real time of 2 days vs 2 hrs