Heterotrophic Bacterial Enumeration

Dear to Whom it may concern,

I am conducting research with my professor, Dr. Al Mikell, at Oklahoma Christian University in reference to total heterotrophic bacteria examination in non-potable pond waters for our water filtration research project. Our particular research is geared to making water filtration cheaper and more effective for third world countries. The specification of my role within this project is examination of variation of diluents in reference to dilution efficacy comparable to standards. Beyond actually counting of cells , we would like to examine colony morphology and physical characteristics for qualitative and quantitative comparisons of varying CFUs (colony forming units). I examined the various pipelines and had difficulty in actually being able to create a software that could have all of the needed specifications. I saw that the yeast patch and yeast colonies examples given via your webstite, and they were similar to what I need if they were combined together. I was at a loss to come to conclusion of the method to do so, however, thus posting my concern here. I also am trying to be able to differentiate within a higher variation of colors (purples,blues,greens,yellow,whites, oranges) than those used within these examples. There are other traits that determine the colony morphology such as surface gleen, opacity, and shape, which would be helpful in determination, but understood if unable to differentiate within this program. I have attached two picture of a couple of the plates that I am attempting to be able to count. Thank you for reading this and considering to help. Hope you have a great day!

Hi Clayton,

The analysis problem you’ve posed is fairly challenging, in that you have a wide variety of objects on a dish with heterogeneous illumination. I’m attaching a pipeline plus a binary mask file that attempts to find objects on the IMG_3242 image. However, some caveats:

  • The binary mask only works for the one image because the other image was acquired with a different camera/dish position. In order to do this in an automated fashion, you need to have each dish in a consistent position/distance so that the same mask can be used for all of them.

  • Hypothetically, a mask could be created for each of the 10 regions which may make things easier. However, again, this would need to be spatially consistent/precise from dish to dish.

  • From seeing projects similar to this, one issue that often arises is that the dish is inconsistently illuminating from sample to sample, making automation problematic. As above, I recommend making the illumination as consistent as possible as well.

Based on the colony shown here, you probably won’t be able to classify your colonies with CellProfiler alone. However, you can

  • Use CellProfiler on your images to identify the colonies, measure as many features as you can, and write the per-colony measurements to a file.

  • Use the machine learning tool in CellProfiler Analyst (CPA, available here
    ) to allow a user to visually sort a “training set” of colonies of various phenotypes, let the computer come up with the criteria to discriminate between them using the measurements from above, and then score all the cells in the experiment. This will also allow you to sort out the objects which are not colonies at all, such as false positives produced by the pipeline.

Once the CP analysis of the images, is to your liking, then you can give CPA a try. A CPA forum is also available for your use. I hope this helps get you started!


2011_02_23.cp (9.58 KB)