###########################################
###########################################
- OVERVIEW
- PACKAGE REQUIREMENTS
- DATA REQUIREMENTS
- USAGE
- USER VARIABLES
- ACCREDITATION AND LICENSE
###########################################
###########################################
Code designed to segment images of fluorescent E. coli cells and produce usable time series of counts of both planktonic cells and cell aggregates. A video showing image segmentation - prior to filtering of cell aggregates - is linked below:
Below is a brief overview of the segmentation process for planktonic cells, which can be near each other and potentially overlapping:
(a) A region of interest containing possible cells is extracted from the initial, larger image (this example chosen as it was already known to contain three real cells).
(b) The region is split up into 14 binary masks representing regions of varying brightness, which are eroded then dilated for simplification then re-layered into a 15-color composite mask.
(c) Regions of locally maximum brightness are identified in the composite image and chosen as seed locations for generating possible cell boundaries.
(d) The seed regions expand to cover the composite image (excluding the background), representing possible cell boundaries which can be projected back onto the original region of interest to generate information later fed into an SVM to be classified as cells or noise.
Full details, including an overview of image quality assessment and time series generation, are available in our paper:
Quantitative high-throughput population dynamics in continuous-culture by automated microscopy
The latest version of the code is fully commented and contains short summaries at the beginning of each Python script (located in the 'scripts' folder).
###########################################
###########################################
- Python 2.7
- NumPy
- SciPy
- scikit-learn
- scikit-image
- (NOTE: Code has only been tested under Ubuntu 14.04/16.04)
###########################################
###########################################
Images are assumed to be 0-255 grayscale and saved at 1288x964 in the .pgm format, with the filename convention PREFIX-aaaaaa-bb.pgm, where "PREFIX" is any string, "aaaaaa" is the 6-digit number of the minute the image was taken during, and "bb" is the 2-digit frame number of the image within that minute. Therefore the first image in an image set might be "PREFIX-000000-00.pgm".
NOTE: Although the above filename convention implies up to 100 frames can be taken per minute, the code is set up under the assumption that there are only 5. Image sets containing more than 5 frames per minute would require modification to many of the analysis scripts.
###########################################
###########################################
After setting the user variables defined below, running "automatic_segmentation.py" from this directory will automatically produce time series for cells and clumps detected in the images. Data is saved to a folder with the experiment name specified by the user variable "expLabel", with cell counts in "expLabel_counts.csv" and clump counts in "expLabel_clump_counts.csv". In both files the first column is time in minutes (as specified in the image filenames), the second column are counts, and for cell counts a third column with a standard error of counts across frames is also included. Note that only minutes determined to include good cell count data are included, so some minutes may be missing from the .csv file despite being included in the images.
In practice, the analysis scripts here do not always correctly eliminate bad time points (such as those dominated by large bubbles). If this is a concern, the script "manual_segmentation_part_1.py" should be run first. This will generate a tentative "expLabel_counts.csv" file, which can be analyzed by hand (or a new script) and should be saved in place with any necessary deletions. Next, the script "manual_segmentation_part_2.py" should be run, using the same experiment label as in the first script.
###########################################
###########################################
The scripts are currently set up to be used on the test images included. Alternatively, all user variables should be edited by hand in whichever of the following scripts are used:
- automatic_segmentation.py
- manual_segmentation_part_1.py
- manual_segmentation_part_2.py
The user variables are:
- expLabel: A descriptive label for the experiment. Determines the name of the folders and files generated. If left blank, defaults to the date the script was run.
- imgDirectory: The directory containing the images.
- filePrefix: The image filename prefix. Note that where filePrefix is PREFIX, the assumed filename convention is PREFIX-aaaaaa-bb.pgm, where "aaaaaa" is the 6-digit minute number the image was taken during and "bb" is the image's 2-digit frame number (note that the code currently requires there to be 5 frames per minute). The first image of an experiment might be PREFIX-000000-00.pgm (the test images included are from later time points).
- firstMinute: Usually would be 0; note that this corresponds to the lowest value of "aaaaaa" of any image file according to the convention defined above.
- experimentLength: Usually the number of minutes in the experiment; note that this corresponds to the highest value of "aaaaaa" of any image file according to the convention defined above.
- trainingSetFilePath: The filepath of the machine learning training set for this strain and experimental condition.
- clumpBrightnessThreshold: A clump must have a max pixel intensity (after smoothing) of at least this value or be discarded. The intensity range is 0-255.
- clumpMeanBrightnessThreshold: A clump must have a mean pixel intensity (after smoothing) of at least this value or be discarded. The intensity range is 0-255.
- clumpSizeThreshold: A clump must have a pixel value area at least this high or be discarded. Note that the minimum clump size is already 500 pixels internally; clumps must have an area of at least 500 pixels to be counted regardless of this threshold value.
###########################################
###########################################
The code for the "maskhist" method in the imseg.py script is a slightly modified version of the "histogram" function from the scikit-image exposure package (skimage.exposure).
The code for the "Moments" method in the imseg.py script was ported to Python from a port by Gabriel Landini (for the Auto_Threshold plugin of the Fiji project for ImageJ) of code written by M. Emre Celebi for the open source project FOURIER: http://fiji.sc/Auto_Threshold#Installation https://github.com/fiji/Auto_Threshold/blob/master/src/main/java/fiji/threshold/Auto_Threshold.java http://www.lsus.edu/faculty/~ecelebi/fourier.htm http://sourceforge.net/projects/fourier-ipal
Other code written by Jason Merritt and Seppe Kuehn at the University of Illinois at Urbana-Champaign.
This code is covered by the GPLv2 license, the text of which is included in the LICENSE file.