RBC and WBC Blood Cell Counter using Circular Hough Transform 

Here are the specific steps followed for cleaning and pre-processing the dataset:

• Load blood sample image using imread function

• Convert the given image to gray image using rgb2gray function

• Increase the contrast of the image using adapthisteq function (optional)

The following steps can be used to eliminate platelets and other smaller objects from the image and to segment RBC and WBC.

• Use bwareaopen command to reduce noise components such as platelets and other small dots (generally salt and pepper noise).

• Binary inversion of the image to get Blood cells as white (are of interest) and background as black (BW2= ~BW1, where BW2 is a binary inverted image of BW1)

• Fill the holes using the imfill command (optional)
• Use bwareaopen command once again to eliminate all the smaller objects from the image (optional)
• Use imerode function with a disk-shaped structuring element called strel , to separate the touching border of the cells.

Following these steps, you get an image containing only RBC and WBC!

Once the WBCs and RBC’s are segmented we can easily separate them, as the size of WBCs is bigger than RBCs.

• The stats = regionprops('table',Segmentedimage,'Area'); command can be used to find the area for separating WBC and RBC. This shows the area of all the RBCs and WBCs involved in the image.
• Once the area of WBC’s is known they can be separated using the command BW2 = bwareafilt(BW,range), where the range is from the lowest to highest WBC areas. This will extract all the WBC cells within that area range from the image.

• Fill the holes using imfill command (if required)

• Use the bwperim function to obtain the edges of the segmented WBC cells and overlay this edge mask on the image.

• Use bwboundaries or bwconncomp functions to find connected components in binary image. (The number of connected components is same as the WBC count present in the sample).

Follow the same steps as specified in the WBC segmentation; the only difference here is the area range used for RBC segmentation.

• The stats = regionprops('table',Segmentedimage,'Area'); command can be used to find the area for separating WBC and RBC. This shows the area of all the RBCs and WBCs involved in the image.
• Once the area of RBC’s is known they can be separated using the command BW2 = bwareafilt(BW,range), where the range is from the lowest to highest RBC areas. This will extract all the RBC cells within that area range from the image

• Fill the holes using imfill command (if required)

One of the problems encountered while counting the RBC’s is due to overlapping. The cells will appear as a single cell when they overlap each other. So sometimes even for a lab technician, it becomes difficult to differentiate. So manual counting is more tedious and error-prone.
In order to count the overlapped cells in the RBC, we need to consider the fact that RBC’s are having a shape similar to circular shape and WBC’s are irregular in shape. So here in this method, we are using a circular Hough transform, which is an algorithm used to identify the circular objects in an image. The accuracy of this identification will be reduced when more than 4 cells are overlapping. [NOTE: A good blood sample has not more than three cells overlapping].

• Use the bwperim function to obtain the edges of the segmented RBC cells and overlay this edge mask on the image

• Apply imfindcircles function to perform circle Hough transform and use length(centres) to get the RBC count.

             [centres, radii, metric] = imfindcircles(A, [Rmin, Rmax],…
            'ObjectPolarity','bright','Sensitivity',0.9,'EdgeThreshold',0.1);

Here A is the input image and Rmin and Rmax are the radius range of the circular objects or in other terms the RBC’s you want to detect. Specify a relatively small radius range for better accuracy. In this code, we have used [20, 50] in place of [Rmin, Rmax].
Object polarity indicates whether the circular objects are brighter or darker than the background, specified as the comma-separated pair consisting of 'ObjectPolarity' and either of ‘bright’ or ‘dark’. Here in this application RBC’s are brighter than the background. Therefore it’s specified as 'ObjectPolarity','bright'.
Sensitivity specified as the comma-separated pair consisting of 'Sensitivity' and a number in the range [0,1]. As you increase the sensitivity factor, imfindcircles detects more circular objects, including weak and partially obscured circles. Higher sensitivity values also increase the risk of false detection. Since RBCs are not perfect circles and having a biconcave disk shape it's highly recommended to use a higher sensitivity value of 0.9.
imfindcircles detects more circular objects with both weak and strong edges when you set the edge threshold to a lower value. It detects fewer circles with weak edges as you increase the value of the threshold. So in order to count all the RBC’s present in the sample, it’s advisable to use a lower edge threshold value of 0.1. The accuracy of detection of overlapped blood cells is higher when we precisely tune these input parameters.
The circle centers and its corresponding radius values obtained using this algorithm is stored into the output parameters called centers and radii. Therefore the length of centers will give you the count of overlapped RBC’s present in the sample.