Methods and Testing

In the initial phase of our project, we interviewed a microbiologist at bioMeriéux to learn about common experiences and preferences for an automated bacterial colony counting system. This process allowed us to obtain the “voice of the customer,” which is central to our goal of creating something that has tangible value to scientists today.  Incorporating the “voice of the customer” is what separates this project from the other automated image processing-based colony counters that are available today. After the interview, we researched current open-source solutions to this problem to obtain new perspectives from other approaches to the problem, as well as to evaluate their shortcomings.
Multiple MATLab scripts were then written utilizing the Hough Transform within the Image Processing Toolbox’s imfindcircles function. Overall, these scripts import an image, convert it to a grayscale image, and using the Hough Transform with specified radii, contrast, and sensitivity settings, work to detect the number of bacterial colonies present.

The specifics on each method can be found by following the links to the right, under ‘Methods’.

We then began initial testing of our program, starting with images (found online) of sparsely populated agar plates where the colonies can be counted easily by hand. As we worked the bugs out of our code, we graduated to denser, more heavily colonized plates to run through and test. This allows us to not only determine the code’s accuracy, but also to find what additions can be made to make the code more complex and allow it to most optimally run through different bacterial colony patterns and colors. After making these initial edits to our code to accommodate the small sample size variation in images found online, we were ready to progress to acquiring images of our own and running them through our algorithms; this way we can test the entire process from image acquisition through colony counting while employing filtering intermediately.

An apparatus was built to aid in image acquisition, shown below in Figure 7. It is made from mixed materials, due to the limited quantity of scraps available in the Wash U machine shops. It is composed of three parts: a top camera mount made from a delrin plastic, a back stand made from steel, and a rectangular base made from aluminum. Each side of the top camera mount fits one of the two camera models that we were given for use in this project (Point Grey Blackfly and Edmund Optics Monochrome); in other words, the mount is reversible to allow for interfacing with either camera model. The back stand is three feet long, to account for the maximum focus length between the two cameras, and fits snugly and securely into a rectangular indentation in the top camera mount, as well as a similar indentation in the base. The base is just large enough to accommodate a standard agar plate. It was originally planned to create a circular indentation in the base to place the plate to be imaged; however, due to the less than ideal materials available, it was decided that this indentation would decrease the weight of the base, and jeopardize the stability of the apparatus.

Figure 7: Image acquisition apparatus, shown interfaced with laptop with live image of plate on screen.

CAD drawing of our setup

 

CAD detail view of the reversible camera mount (top piece), usable with two different cameras: The Edmund Optics 0312M (red) and the Point Grey Blackfly (grey).

Before we were able to begin testing on our program, bacteria needed to be grown in plates. The plating and incubation was done for us by Brad Clay, our Industry Mentor. Once the bacteria had incubated for the required 24 hour period, the plates were able to be removed from the incubator, and then imaged using our image acquisition apparatus.

The first set of images was taken using experimental dilutions of the bacteria strains, which ended up producing extremely heavily populated dishes. This showed us that we needed to do additional dilutions for future imaging. However, images of these plates were still taken and used to run through initial stages of our algorithms to get it to a more advanced and reliable stage.
Once this was done, a second set of images was acquired using further dilutions, this time using varying backgrounds and light intensities. The backgrounds used were: the natural aluminum background of the apparatus, a white background, and a black background. The room available to us for imaging the agar plates had direct overhead lighting, which caused reflections on the agar and the colonies themselves. Images were taken of the plates with this direct lighting, as well as with a shield above the plate to shade it from the direct light, in hopes of minimizing any reflections. Images were acquired featuring combinations of all backgrounds and light intensities, to create a comprehensive set of 140 images. These images were then used to run through our algorithms for each method, all while making appropriate edits to compensate for variables that were not originally anticipated.