Source-linked AI summary
OpenCFU, a New Free and Open-Source Software to Count Cell Colonies and Other Circular Objects
Quentin Geissmann
TL;DR
Biological counting is important but remains predominantly manual because it is time-consuming and subjective. This paper introduces OpenCFU, an open-source tool using robust image processing and a modern interface, and reports that it is faster, more accurate, and more robust than two tested alternatives. The study also positions it for counting other circular biological objects.
Problem
Counting biological objects remains predominantly manual despite being time-consuming and subjective.
Method
OpenCFU combines recursive multi-threshold circular-region analysis with an open-source, cross-platform interface for processing images and video streams.
Results
OpenCFU was faster, more accurate, and more robust to common perturbations than NICE and Cai et al.’s macro in the tested conditions.
Takeaways & Limitations
OpenCFU can count bacterial colonies and other circular objects such as seeds, pollen, cells, nuclei, and eggs without software modification.
Takeaways & Limitations
The study could not formally compare OpenCFU with additional software tools, including Clono-counter, for different reasons.
Abstract
from arXiv · showhide
Counting circular objects such as cell colonies is an important source of information for biologists. Although this task is often time-consuming and subjective, it is still predominantly performed manually. The aim of the present work is to provide a new tool to enumerate circular objects from digital pictures and video streams. Here, I demonstrate that the created program, OpenCFU, is very robust, accurate and fast. In addition, it provides control over the processing parameters and is implemented in an in- tuitive and modern interface. OpenCFU is a cross-platform and open-source software freely available at http://opencfu.sourceforge.net.
Introduction
Biological object counting is important but remains commonly manual because it is time-consuming, subjective, and technically challenging. OpenCFU is introduced as an open-source tool designed to improve processing performance, accuracy, robustness, usability, and applicability.
- Motivation: Biologists routinely enumerate objects such as pollen, eggs, seeds, nuclei, cells, and organisms.
- Motivation: Colony counting requires background separation, overlap handling, and rejection of artefacts such as agar imperfections, dust, and Petri-dish edges.
- Motivation: Existing tools have seen limited adoption because of performance shortcomings, insufficient robustness, difficult interfaces, and limited public availability.
- Contribution: OpenCFU was developed as an alternative open-source tool with robust, accurate, and fast image processing plus a modern, functional interface.
- Contribution: The study compares OpenCFU with NICE and Cai et al.’s ImageJ macro, reporting that OpenCFU is faster, more accurate, and more robust to usual artefacts.
Algorithm and Implementation
OpenCFU increases robustness by testing circular-region validity across multiple thresholds, then filtering and classifying candidate components. Its implementation supports rapid, interactive analysis and structured colony outputs.
- Algorithm: The algorithm tests multiple grey-scale thresholds and retains regions that are recurrently morphologically valid, reducing false positives from high-contrast artefacts.
- Algorithm: A particle filter evaluates area, perimeter, convexity, aspect ratio, and hollowness to identify regions likely to contain circular objects.
- Algorithm: A score-map records how recurrently pixels belong to circular regions, after which thresholding and a second particle filter classify components as invalid, individual, or multiple objects.
- Algorithm: An optional normal-distribution model uses relative colour intensities to exclude marginal unsplit objects through a user-controlled likelihood threshold.
- Implementation: OpenCFU uses C++, OpenCV, OpenMP, and GTKmm to provide cross-platform processing and a graphical interface.
- Implementation: When parameters change, the software restarts only from the first affected processing step rather than reanalysing the image from the beginning.
- Implementation: Outputs include image-level colony counts and detailed per-colony measurements such as position, colour intensity, area, and perimeter.
Results
Across tested images and plates, OpenCFU was fastest, showed no significant counting bias, and approached human accuracy. It also appeared more resistant than NICE and IJM to common artefacts and acquisition perturbations.
- Speed: O(n) processing was observed for all three algorithms, with OpenCFU fastest, followed by IJM and NICE.For a typical 1.6×1.6kpx picture, processing times were approximately 0.69, 1.22, and 3.0 seconds, respectively.
- Accuracy: The accuracy experiment analysed 19 Staphylococcus aureus plates containing between 10 and 1000 colonies, using seven trained human counts as reference.
- Accuracy: OpenCFU showed no significant counting bias, with regression slope a = −2.58% and P-value = 0.169.
- Accuracy: NICE and IJM significantly exceeded human inaccuracy, whereas OpenCFU’s median error of 1.93% was not significantly less accurate than humans.
- Image quality: With poor-quality images, median errors increased to 2.78% for OpenCFU, 13.0% for NICE, and 11.1% for IJM.
- Robustness: NICE and IJM appeared prone to falsely counting bubbles, edges, cracks, and dust, whereas OpenCFU appeared unaffected.
- Robustness: A 25-pixel translation simulated 1.7mm dish mispositioning to quantify sensitivity to acquisition-edge perturbations.
- Robustness: Bubble-only plates were used to test whether detected-colony counts increased with bubble number across methods.
Discussion
OpenCFU combines fast processing with accuracy and robustness to artefacts, while supporting low-cost acquisition and practical postprocessing. Its comparisons also identify limits in evaluating alternative tools and show that verification can reduce their apparent time savings.
- Limitations: Formal comparison was limited because some additional tools were unavailable, unsuitable for large or numerous images, or could not be adapted to the pipeline.The study therefore compared OpenCFU with two available alternatives rather than the full set of related tools.
- Performance: OpenCFU generated no more errors than average human counting error on high-definition pictures, whereas NICE and IJM were significantly less accurate.The alternatives overestimated plates with fewer than 50 colonies and underestimated plates with large numbers of colonies.
- Practical use: A low-cost webcam combined with OpenCFU provided very satisfying results, and the software integrates webcams, USB microscopes, and firewire cameras.This supports image acquisition without relying exclusively on slower desktop scanners.
- Robustness: OpenCFU was not affected by bubbles and remained robust when acquisition conditions changed slightly, unlike the alternatives.The discussion highlights dish mispositioning of less than two millimetres as a perturbation that could substantially inflate alternative-method counts.
- Practical use: Although the three methods could process 19 plates in under five minutes, alternative-method verification and correction could negate their time gain.Humans took 39.3 (sd=6.1) minutes to count the 19 plates; OpenCFU’s rapid display is also useful during calibration.
- Practical use: OpenCFU’s postprocessing filters remove detected objects by intensity and colour similarity, including contaminant bacteria and writing.Detailed output also provides information about each detected object, supporting analyses of distinct cell populations.
- Performance: OpenCFU was faster, more accurate, and more robust to common perturbations than the two tested alternatives under the tested conditions.The study reports these properties across processing speed, counting accuracy, artefact resistance, and positioning perturbations.
Availability and Future Directions
OpenCFU is freely available as open-source software with supporting documentation and sample results. Future updates are planned in response to user feedback and include broader region selection, outlier exclusion, and command-line use.
- Availability: OpenCFU is distributed under the GNU General Public Licence version 3 and is available on Sourceforge.A user manual and video tutorial are also provided.
- Availability: An expanding collection of image samples and corresponding results will help users calibrate the method.The collection is intended to support practical use of the software.
- Future directions: The program will be improved and updated as users request features and report bugs.The paper presents development as responsive to user feedback.
- Future directions: Planned improvements include multiple regions of interest, supervised exclusion of outlier colonies, and a command-line version.These additions target broader analysis control and automation.
Materials and Methods
OpenCFU combines multi-threshold image processing, particle filtering, and watershed-based segmentation to identify individual and clustered circular objects. Accuracy was assessed against blinded human counts using standardized bacterial plates and statistical comparisons across agents and image conditions.
- Particle filtering: Particle filtering rejects regions outside user-defined radius criteria and distinguishes single objects from clusters using separate acceptance conditions.The maximal radius may be user-defined or calculated from image dimensions, while clustered objects must satisfy aspect-ratio, width, and area constraints.
- Image-processing pipeline: During the second pass, connected components are classified as invalid, individual objects, or multiple objects; clustered objects are segmented with a distance-map watershed variant.Individual objects are accepted, invalid objects rejected, and multiple objects processed morphologically.
- Image-processing pipeline: Distance-transform peaks provide watershed markers, while region growth is limited by D = 1.6 × r, where r is the presumptive radius from the corresponding peak.The distance transform uses a 5×5 pixel neighbourhood with a=1, b=1.4, and c=2.1969.
- Statistical analysis: Agent deviations were calculated relative to each plate’s median human count, with regression, t-tests, and Wilcoxon tests used for specified comparisons.Analyses included effects of colony number, image translation, and bubbles, using R software.
Figure Legends
Figures 1–6 depict OpenCFU’s processing pipeline and evaluations of its speed, accuracy, robustness, and versatility across images and circular biological objects.
- Processing pipeline: Figure 1 outlines preprocessing, score-map generation, connected-component detection, watershed segmentation, and optional postprocessing filters.Preprocessing corrects gradual background changes and increases contrast; postprocessing may be performed by OpenCFU or the user.
- Processing pipeline: Figure 2 illustrates channel-wise preprocessing, score-map thresholding, watershed separation of merged colonies, and intensity or colour filtering.Arrows identify morphologically unlikely regions, successfully segmented objects, excluded contaminants, and valid colonies.
- Performance evaluation: Figure 3 compares processing time for OpenCFU, NICE, and an ImageJ macro on images varying in colony number or colony size.OpenCFU is reported as faster than both alternatives over the tested resolution range.
- Performance evaluation: Figure 4 compares method accuracy against median counts from seven humans, examining colony-number effects and absolute deviation under high- and low-definition imaging.OpenCFU did not show the significant bias observed for NICE and the ImageJ macro, while low-definition images increased error for all three methods.
- Robustness and versatility: Figure 5 evaluates robustness to bubbles, agar cracks, dust, dish edges, plate mispositioning, and bubble-only plates using qualitative and quantitative comparisons.The figure reports artefact detections, translation-induced bias, and relationships between bubble number and detected objects.
- Robustness and versatility: Figure 6 demonstrates OpenCFU on bacterial colonies, mustard and soy-bean seeds, and Carduus pollen across clear, poor-quality, low-contrast, and noisy images.The examples assess versatility using representative 200×200-pixel regions.