2 A M ATL AB A pproach to Automated Bacterial Colony Counting Jackie Marlin #1*, Jake Starr# 2* #D epartment of Electrical and Systems Engineering Washington University in St. Louis 1 Brookings Drive St. Louis, MO 63130 USA 1j [email protected] 2j [email protected] *These two authors contributed equally to the work Abstract The purpose of this project is to use M ATL AB to create an automated method to count the number of bacterial colonies on a given agar plate, from an image of the plate. Counting the number of colonies on an agar plate is an extremely common activity in microbiology labs for medical, pharmacological, and agricultural uses. Currently, colony counting is done mostly manually because devices made for automatic counting are expensive and not user­friendly. For this project, the Circular Hough Transform is utilized to detect circular features from the images. These images were taken under differing lighting conditions for 20 combinations of agar types and species of bacteria; this provided a large dataset to develop our four different algorithms. Our M ATL A B script is implemented to ultimately output a colony count with up to 98% accuracy, depending on the quality of the original image. Index Terms— Biology computing, Edge detection, Image analysis, Image capture, Image quality, Object detection noise or vibration from the apparatus that verifies the counting of each tapped colony. The device stores the total number of I. IN TRODUCTION taps as the number of colonies present on the plate. This NE of the most common tasks that takes place in process is still quite time consuming, and only slightly microbiology labs is the counting of bacterial eliminates human error. Additionally, it still takes colonies that grow on agar plates. The most basic, and microbiologists away from more skilled tasks. A typical common, way to accomplish this task is by manually manual colony counter is pictured in Figure 2. counting each colony that is present on the plate with a The most advanced labs may utilize automated colony clicker, shown in Figure 1. However, this is extremely time counters, but these tend to be extremely expensive, consuming, requires little thought from the highly trained error­prone, and user­unfriendly [3]. These automated microbiologists, and there is significant room for human colony counters operate by capturing an image of the error. plate, and then using various image analysis techniques, A slightly more advanced manual counting technology such as altering contrast values and edge detection, process utilizes a hatched grid and a light source (overhead or the image and count the number of colonies present. A below) that holds a transparent agar plate. Moving square commercial example of an automated colony counter is by square on the hatched grid, the microbiologist will tap shown in Figure 3. each individual colony with a stylus. Each tap signals a Testimonials from current microbiologists suggest that these automated counters are far from ideal, yet clearly 3 neither is manual counting. There is significant need for a used on images of agar plates with bacteria growth present bacterial colony counting method that is affordable, to most accurately determine the colony count were accurate, and efficient, and this project works towards the researched. development of such a method [5]. Past research shows that the main obstacles in creating an automated colony counter are illumination (shadows, reflections), bacterial colony overlap, and imperfections in the agar or plate itself (dust particles, lumps, etc). However, colony overlap and imperfections can be tackled using image binary conversion, where all pixels less than a set grayscale value are deleted from the image [7]. This eliminates potentially problematic pixels that could interfere with the counting and introduce false positives. Figure 1 (above): A clicker and a marker used to manually count bacterial colonies on an agar plate [1]. Figure 3 (above): Automated colony counter, with output interface shown behind. Captures image of plate, analyzes and filters image, and returns the number of colonies as well as an image of what the device determined to be colonies [4]. Our program implements the Hough Transform, which Paul Hough developed in 1960 as a “method and means for recognizing complex patterns” [12]. Hough’s intention was to streamline the analysis of bubble chamber pictures since many photographs needed analysis and when done by hand each photograph took several hours to analyze Figure 2: A more advanced manual colony counter with an attached [12]. Essentially, Hough was solving the same problem for overhead light. The circular region contains a hatched grid on which a the study of atomic particles as we would like to solve for stylus is tapped on each visible colony. The machine registers each tap and stores a count that will ultimately be the number of bacterial colonies the calculation of sample bacterial concentrations. In present [2]. MA TL AB, the circular Hough Transform works by first analyzing the given image for high­gradient pixels ­­ that is, pixels that are a vastly different color than the A. Review of Literature surrounding pixels. By selecting several of these high­gradient pixels in close proximity to one another, it In addition to research regarding currently available can be seen how they are related to one another. In our automated colony counters, the various filtering methods case, we are analyzing only for a circular relationship 4 where the given high­gradient pixels lie on the edge of a can cope with a high range of bacteria­agar contrasts, and common circle. Based on the angle and distance between will produce a good­sized sample group of images for our the points, the program can estimate the radius and center initial testing. of a found circle [13]. The circular Hough Transform is a good option to use due to its robustness in the face of image noise and its ability to detect even circles that are only partially visible (which is important for the detection of intergrown colonies). B. Aims and Objectives This project aims to optimize bacterial colony counters to: □ Overcome some of their main inconveniences, and □ Incorporate some requests for improvement directly from professional microbiologists who use these current models of colony counters on a daily basis. Figure 4: The different colors and transparencies of agar to be used, from left to right: (top) SabDex, TSA, Mueller Hinton, (bottom) TSA­B, and Input from professional microbiologists, or the “voice of Chocolate Agar. Each agar works with different bacteria types, which the customer” input, will ensure that our project will be allows for a variety of contrast combinations to test with the code. Source: useful to lab technicians and microbiologists today, as well Own image. as demonstrate the necessity for the development of such a tool. II. ME THODS AND P R OCEDURES This project will be completed using different In the initial phase of our project, we interviewed a processing methods available within the MA TL AB Imaging microbiologist at bioMeriéux to learn about common Toolbox, interfaced with cameras ranging from 1­5 experiences and preferences for an automated bacterial megapixels provided by bioMeriéux, ultimately to be used colony counting system. This process allowed us to obtain to capture images of colonies of bacterial strains differing the “voice of the customer,” which is central to our goal of by size, color, and density. Specifically, we will be testing creating something that has tangible value to scientists five different bacterial species: E. Coli (Gram Negative), today. Incorporating the “voice of the customer” is what Klebsiella Pneumoniae (Gram Negative), Staphylococcus separates this project from the other automated image Aureus (Gram Positive), Pseudomonas Fluorescens (Gram processing­based colony counters that are available today. Negative), and E. Fissicatena (Gram Negative). In addition After the interview, we researched current open­source to different bacteria, different types and colors of agar will solutions to this problem to obtain new perspectives from be used: SabDex Agar, Tryptic Soy Agar (TSA), Mueller other approaches to the problem, as well as to evaluate Hinton Agar, Blood Agar (also known as TSA­B), and their shortcomings. Chocolate Agar, pictured below in Figure 4. SabDex is Multiple MA TL AB scripts were then written utilizing used for a variety of Gram­positive and ­negative bacteria the Hough Transform within the Image Processing and fungi, and is composed largely of Dextrose [5]. TSA is Toolbox’s i mfindcircles function. The function takes also largely composed of Dextrose, which is ingested by in several input parameters: bacteria as a main energy source, but is primarily used to □ An image file, isolate organisms from our body. Mueller Hinton is used □ A r adius parameter given as a 1x2 array primarily for antibiotic susceptibility testing of containing both the minimum and maximum radii nonfastidious organisms, or those that are not restricted to of circles to search the image file for, ingesting a particular substance. Blood Agar contains □ An O bjectPolarity parameter specifying about 5% sheep blood and is used to determine the types whether the circles are bright or dark with respect of hemolysis present (the process by which red blood cells to the agar, and are ruptured). Chocolate agar is generally used to grow □ A S ensitivity parameter which sets how fastidious respiratory bacteria, such as flu viruses [6]. circular an object has to be in order to be detected These choices provide a variety of pigmentations and (S ensitivity values are on a scale from 0 to bacterial morphologies to facilitate building a program that 5 1, with 0 and 1 corresponding to “detect only perfect circles” and “detect anything mildly circular,” respectively). The i mfindcircles function returns a matrix containing the radius and (x,y)­center of each circle (colony, in our case) that was detected.
Details
-
File Typepdf
-
Upload Time-
-
Content LanguagesEnglish
-
Upload UserAnonymous/Not logged-in
-
File Pages11 Page
-
File Size-