What MyoVision did

Algorithm worflow

When you ran the example, MyoVision:

• opened a configuration file (configuration_data.xml) stored in ..\demos\example_A\configuration_data.
  • looked at the task file section
    • saw one task specified with
      • an input image file PoWer_3_Gastroc_10x_blue_cropped_small.png stored in ..\demos\example_A\raw_image
      • results to be written to ..\demos\example_A\results

• ran the task by:

  1. reading in the input image file

     

  2. converting it to gray-scale

     

  3. saturating the image
     • this step adjusts the brightness and contrast of the gray-scale image until x percent of the pixels are as dark as they can be, and x percent are as light as they can be
     • the x value, in this case 10%, is set by the saturation percent variable in the <image_to_label_parameters> of the configuration file

     

  4. enhancing the cell edges using a Frangi filter

  5. performing an initial segmentation to identify ‘blobs’ - regions of connected pixels
     • discarding blobs with an area below a critical size
     • the critical size, in this case 50 pixels, is set by the min_object_size in the <image_to_label_parameters> of the configuration file

     

  6. loading a classifier model (more on this later) set by classification_model_file_string in the <classifier_parameters> section of the configuration file
  7. using the classifier to mark each blob as one of the following:
     • fiber
     • potentially-connected fibers (two or more fibers separated by an incomplete cell boundary)
     • something else

     

  8. attempting to separate potentially-connected fibers into single fibers
     • this is done using a watershed algorithm and a watershed_distance_parameter (in this case, 10 pixels) defined in the <image_to_label_parameters> section of the configuration file

     
     

  9. refining the edges of each cell using an adaptive contour technique which depends on the following parameters which are defined in the <refine_fibers_parameters> section of the configuration file:
     • max_iterations
     • sigma

     
     
     

  10. re-segmenting the image to identify fibers with improved fidelity

      
      

  11. correcting for fibers that were inappropriately connected during the fiber refinement step

      
      

  12. creating the following output files in results_folder (which was specified for the task in the <task_files> section of the configuration file

      • an Excel file final_results.xlsx containing the following metrics for each fiber:

        • label (the id for the fiber)
        • area (in pixels)
        • eccentricity (0 is a circle, 1 is a straight line)
        • convex area (in pixels)
        • equivalent diameter (in pixels)
        • extent (ratio of pixels in the blob to the pixels in the total bounding box)
        • major axis length (in pixels)
        • minor axis length (in pixels)
        • orientation (angle between horizontal and the major axis of the equivalent ellipse - ranges from -pi/2 to +pi/2)
        • perimeter (in pixels)
        • solidity (ratio of pixels in the blob to the pixels in the convex hull image)

      • an image file clean_overlay.png showing falsely-colored fibers overlaid on the original input image

      • an image file annotated_overlay.png showing falsely-colored fibers overlaid on the original input image with text labels showing the fiber IDs

      • a zip file processing.zip which can be useful for trouble-shooting and which contains:

        • a folder named blocks with images showing processing steps for individual blocks
        • a folder named classification_steps with images showing the results of the processing steps outlined above