MORPHE Files

Copyright (c) 2016, Tzu-Yu J. Liu, Yun S. Song
All rights reserved.


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% MORPHE %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%


One can also use the software as a feature extraction and visualization 
toolbox and run further analysis using the extracted information. To use 
the software, create a folder for the colony images of the same class, and 
create two subfolders “GFP” and “RFP”. The GFP folder should contain the 
images of GFP fluorescence, each ends in “_c1.tif”. The RFP folder should 
contain the images of RFP fluorescence, each ends in “_c2.tif” and 
corresponds to a GFP image with the same prefix. An example can be found 
in the "data" folder. A folder named “statistics” will be created after 
the analysis, in which the extracted features in .mat format and images of 
the features overlaid on top of the raw images are saved. 


Run MORPHE.m to start the GUI. The function of each parameter in the 
feature extraction panel is explained below.

(a) file/folder: Select file to analyze a single image; select folder for 
    batch processing.  
(b) Load: A dialog window opens for selecting the file or folder, depending 
    on the specification of file/folder. For exmample, select 
    /data/illustration/GFP/exmample_c1.tif to analyze that image, select 
    /data/illustration to anayze the all the images in the class named 
    illustration.
(c) Clear: Clear the selection. Return to step (a-b)
(d) colony edge detection: Sensitivity thresholds for the Canny edge 
    detection method. The higher the threshold, the fewer colony edge 
    pixels detected.
(e) colony dilation:  The radius (pixels) of a flat, disk-shaped 
    structuring element. The software uses the structuring element to 
    dilate the binary image of (e), such that the detected edges form a 
    close boundary of the colony.
(f) bands/dots edge detection: Sensitivity thresholds for the Canny method. 
    The higher the threshold, the fewer edge pixels of bands/dots detected.
(g) bands/dots dilation: The radius (pixels) of a flat, disk-shaped 
    structuring element. The software uses the structuring element to 
    dilate the binary image of (f), such that the detected edges form a 
    close boundary of each band/dot. 
(h) bands/dots intensity threshold: A number between 0 and 1, denoted as 
    alpha. A local threshold is set as the alpha-th quantile of the pixel 
    intensities of each connected component formed by detected bands/dots 
    in (f-g). Each local threshold is applied to the interior pixels of 
    the corresponding connected component. If the intensity of an interior 
    pixel is above than the threshold, it is labeled as bands/dots.
(i) buffer around bands: The width (pixels) of a buffer around bands to be 
    excluded from analysis. This is useful when there exist large bands 
    with high fluorescence intensity that may create bias in the 
    estimation. 
(j) bands/dots area threshold: The threshold used to divide the detected 
    bands/dots into the class of bands and the class of dots. If the 
    number of pixels of a connected component is above the threshold, that 
    connected component is labeled as a band; otherwise, it is labeled as 
    a dot.
(k) Run feature extraction: Apply the parameters specified in step (d-j) 
    to extract the features. To use the default settings, press the 
    button "Reset parameters".
(l) smoothing kernel bandwidth: The kernel bandwidth applied to the onset 
    frequency of bands and the onset frequency of dots.
(m) marker size: the marker size of the green and blue dots.
(n) Display extracted features: Display the extracted features overlaid 
    on top of the raw images.
(o) To use the default settings, press the button "Reset parameters".
(p) To analyze another file or batch, return to step (a).


The function of each parameter in the visualization and classification 
panel is explained below.
(a) Select/Add classes: A dialog window opens for selecting a folder to 
    classify. Edit the class name of each folder. Folders with the same 
    class name will be treated as the same class. There must be at least 
    two classes specified.
(b) Clear: Clear the selection in step (a). Return to step (a).
(c) smoothing kernel bandwidth: the kernel bandwidth applied to the onset 
    rate of bands and the onset rate of dots. 
(d) Run visualization: Create heatmaps of the smoothed and unsmoothed 
    onset rate of bands and the onset rate of dots. 
(e) Decision tree / Adaboost with decision tree / Random forest: Select 
    the classification method.
(f) Run classification: Use the features smoothed by the kernel bandwidth 
    specified in step (c) to predict the class labels. A heatmap of the 
    confusion matrix will appear after leave-one-out test is done.