Activity 10: Binary Operations

Results:

In this activity our job is to separate a region of interest from the background and determine the average size of a cell to create a method that segregates a cancerous cell from normal cells. Starting from the raw original image in Figure 1. We slice the image into 256 x 256 pixels subimages, the images may overlap, Figure 2 shows the sliced sub images.

Figure 1: Original Images of Cells(punched paper :D)

Figure 2: Sliced Images

After slicing up the images we inspect a sub image and take its histogram. The histogram of pixel values of subimage 1 (topleft portion) showed that setting the threshold value of im2bw to 0.6 most of the background would be turned black. Figure 3 shows the binarized image using im2bw.

Figure 3: Binarized version of subimages

After Binarizing, we intend to separate the blobs from the white dots unfiltered from binarization. Thus we employ the binary operation open which is just an erosion before a dilation. The counterpart of this operation is the closing operation which is just a dilation before an erosion. Using a circular structuring element where the pixel radius is chosen such that the blobs would survive and the background grains would be eliminated. Figure 4 shows the result of the opening operator on the subimages.

Figure 4: Isolation of ROI(cells) after performing an opening operation.

From these processed subimages, the determination of the average cell area is necessary for us to determine the cancerous from the non cancerous cells. We perform a labelling operation using the function bwlabel and determine the histogram of the areas of the image. Since some cells overlap in the image, there are some areas that would be very large. Therefore in taking the histogram of the image we may be able to assert what is the range of the actual cell size. Figure 5 shows the histogram of the areas of all the blobs.

Figure 5: Histogram of Cell Areas

It is clear that most of the cells have areas within 400 to 600 pixels. We can focus on this region and determine the average and standard deviation of the area of a normal cell. Figure 6 shows the focused histogram of the areas.

Figure 6: Focused Histogram from 400 to 600 pixel area.

Cell Area:

Mean = 488.13636

STDEV = 25.527381

We can now use this information in determining if a cell is calcerous or not. Given an image in Figure 7 where there are malignant(Large Blobs) cells included. We must find a way to filter the cancerous ones and discard the normal ones.

Figure 7: Original Image of cells with cancer

It is clear that there are 5 Cancerous cells in the image and just by looking at it we can pinpoint their locations. However, if we use a structuring element that has a radius larger than that of the normal cell or in our case the area covered by the structuring element is larger than that of the average area of the normal cell then we can filter out the normal cells and leave the malignant cells in the image. Upon using this structuring element it can be seen from Figure 8 that the malignant cells are segragated and normal cell filtered out.

Figure 8: Detection of Cancerous Cells.

References:

[1] M.Soriano,"Activity 10: Binary Operations", AP 186, 2010

[2] homepages.inf.ed.ac.uk/rbf/HIPR2/close.ht

Self Evaluation: 10