Detecting Microcalcifications in Digital Mammograms with ImageJ Elizabeth Mobley Visual & Image Processing Summer Ventures in Science & Mathematics Appalachian State University
Abstract The purpose of this project is to show how medical imaging in mammograms detects possible cancerous cells that are unable to be seen to the human eye. This is made possible through a technique called thresholding. We will be using thresholding in the experiment to identify the areas in the mammograms that are cancerous.
Introduction The most common type of cancer that develops in women in western United States is breast cancer. One of nine women will have breast cancer in her lifespan. Before 1969, people took normal x-rays to try and detect the cancer. To detect breast cancer today, radiologists take a new type of x-ray of the breast that is known as a mammogram. [2] Figure 1 shows a mammography machine. The bottom of the arm shows the two plates, and above the plates shows the camera. Mammograms are taken from different angles of both breasts. The mammogram is taken by two plates compressing the breast and then two pictures are taken from each breast. This is done to be sure any of the abnormalities are found with the smallest
amount of radiation needed to take the x-ray. Mammograms are used to help in the aid of diagnosing different types of cancer in the breasts. [1] Figure 2 shows the different clusters of microcalcifications in a normal mammogram in the first image. The second image shows one main cluster of microcalcifications in a breast. Calcifications are what a radiologist looks for in a mammogram. Small clusters of calcium deposits that build up in the breast tissue are known as calcifications. They can sometimes be a warning sign of breast cancer. [1] Macrocalcifications are large clusters of calcifications and are usually not related to cancer. Smaller clusters of microcalcifications will sometimes show in cancerous areas. Depending on the size, shape, and amount of calcifications, the radiologist will decide what to further do with the
images. [3] If cancer is even considered, the doctor may decide to send the images further and have more tests done to the patient to determine if cancer is actually present. The doctor then will look at the images and use different techniques to separate the possible cancerous areas from the non-cancerous areas. The type of technique we will use in this experiment will be thresholding. Figure 3 shows an example of thresholding on an image to make the brighter areas to show more. Using ImageJ, thresholding can be used to show the possible cancerous areas of the breast in the mammogram. The process of thresholding discovers abnormalities in a mammogram. Doctors interpret mammograms by looking for these abnormalities. Some
of theses abnormalities can include asymmetries, irregular parts of increased densities, thick skin, and clusters of microcalcifications. Cancer is shown in a mammogram by bright areas of an image which is also known as high intensity pixels. Thresholding separates the high intensity pixels from the low intensity pixels or dark areas. The thresholding eliminates the lower intensity pixels by setting them to zero to make them black. This allows the cancerous areas of the image to appear clearer and makes a stronger contrast between the non-cancerous areas and the cancerous areas. Radiologist, however, cannot determine whether the patient has cancer or not just by the mammogram. This is because both cancerous and non-cancerous growths in the breast look the same. Methods In this experiment we will separate the high concentrated areas of calcium from the low concentrated regions of calcium. This will show how radiologists predict where the cancer may be located. Using the computer based approach of ImageJ, we can use thresholding to separate the two areas. In the beginning of the experiment, data must be collected of a mammogram. An image of a mammogram must be recovered. The best image to use is a clear mammogram where the shape can be seen to understand the areas that may be cancerous.
Figure 4 shows the original mammogram image that will be used in the experiment. After choosing the mammogram image, thresholding is necessary to further pursue the experiment. The recognized area from thresholding will depend mainly on the level of threshold chosen. The greater the threshold is set to, the less of the image will be shown in the end. We want to see the brightest areas of the image because this is where the most concentrated microcalcifications are located. When setting a low threshold level, more of the image will stay and it will be harder to distinguish the most concentrated areas.
Figure 5 shows the progression of threshold levels. The first image has a threshold value 10. The second image has a threshold value of 50. The third image has a threshold value of 150. Being able to identify the areas of possible cancer is the objective of this experiment. The third image shows these areas the best. Results This experiment shows the progression of higher threshold levels set on a mammogram. The first image with a threshold set to 10 makes it impossible to tell where the most concentrated areas of calcifications are in the white area of the mammogram. This threshold level does not show a good representation of the calcifications because after the thresholding was done, the area highlighted was mainly just the breast area with no sign of calcium dense area. The second image with a threshold set to 50 makes it a little bit
easier to distinguish the area of higher concentration. The third image with a threshold value set to 150 makes most of the areas of high concentration highlighted, making it easy to see the concentrated areas. Conclusion The higher we set the threshold, the better we were able to see the location and shape of the microcalcifications. The higher the threshold was set to, the easier it was to determine where the clusters were located. We concluded that a threshold value of 150 was the best set threshold to determine where the possible cancerous areas were because it minimized the highlighted area to the very bright areas of the image. The ImageJ program will narrow down detected area when we set the threshold to a higher value. The higher the threshold was set, the better results were formed because it minimized the separated area of possible suspected cancer.
References [1] Breast Cancer Organization, http://www.breastcancer.org/testing_mamm_show.html [2] Jacqueline Holland, Detection of Microcalcifications in Digital Mammograms Using Wavelet Lifting Schemes [3] Radiological Society of North America, http://www.radiologyinfo.org/en/info.cfm?pg=mammo&bhcp=1