Análisis de los datos

For the following steps of the registration process, the speed of sound images and the X–ray mammograms have to be segmented into background and ob-

Figure 4.1: This figure shows the histogram of the X–ray mammogram after the segmentation into background and object. The grey value 0 representing the background occurs frequently, whereas the rest of the grey values are normally distributed.

ject. This segmentation is carried out as described in chapter 3.2.2. For the X–ray mammogram the segmentation results in a histogram in which the grey value 0 representing the background occurs frequently, whereas the rest of the grey values are distributed across the entire range of grey values (see figure 4.1).

However, the segmentation of speed of sound images result in a different his- togram. This histogram also has a peak representing the background at its corresponding grey value 0 but the grey values representing the breast are not distributed across the entire range of the grey values. The grey values for the breast range from 1.3 to 1.6 and leave a gap between 0 and 1.3 (see figure 4.2).

This gap has an impact on the projection of the speed of sound images (see chapter 3.3.1) because the variations in the grey values representing the

Figure 4.2: This figure shows the histogram of the speed of sound image after the segmentation into background and object. The grey value 0 representing the background occurs numerously, whereas the rest of the grey values are distributed in a range between 1.3 and 1.6. This leads to a gap between 0 and the grey values representing the object (blue ellipse)

breast are too small in comparison to the difference between background and object. When summating the grey values of a slice of the speed of sound image in order to create a two–dimensional projection, the contrast between breast and background is maintained. However, the contrast between differ- ent tissue structures is not visible in the projection (see figure 4.3).

As tumors have to be visible in the speed of sound image, for example in order to evaluate the quality of the registration, this segmentation needs to be changed to visualize different tissue structures in the projection.

The modification of the segmentation is based on the removal of the “gap” between the grey values of background and object. For each slice of the speed of sound image, the minimum grey value relating to the breast is found. For each grey value of this slice, this minimum grey value is subtracted.

Figure 4.3: The projection of the speed of sound image after the segmentation shows no different tissue structures since the variations in the grey values are too small.

As described in chapter 2.1.2, each slice of the speed of sound images as pro- vided by the Karmanos Cancer Institute has a slightly differing background color. This affects the mean grey value of each slice resulting in a differentia- tion across all of the slices. Hence, every single line can clearly be seen in the projection of the speed of sound image. To create a homogeneous projection (see figure 4.4), the grey values of each slice are divided by the mean grey value of the current slice. These adjustments of the segmentation result in a

Figure 4.4: In the projection of the speed of sound image after the improved segmentation different tissue structures are rendered and the single lines of the projection are marginally visible.

histogram in which the grey values are distributed over the full range of grey values (see figure 4.5). The gap between the grey values of background and object (see figure 4.2) is removed which improves the overall contrast of the

image. Different tissue structures are rendered in the projection of the speed of sound image (see figure 4.4).

Figure 4.5: This figure shows the histogram after the adjustments of the segmentation. The grey values are distributed over the entire range.

The segmentation as described in chapter 3.2.2 was modified in order to en- hance the contrast of the projection of the speed of sound image. In addition, the quality of the projection is improved since the entire projection is imaged more homogeneously. The quality of the projection is crucial for the analysis of the datasets and for the comparison with the X–ray mammogram.

4.3

Datasets

To provide a better comprehension of the following steps, the given datasets are described in this chapter. The three–dimensional speed of sound images are represented by their segmented projections.

Every image shows the breast from right to left in x–direction and from chest wall to nipple in y–direction.

Aside from dataset 1, every X–ray mammogram has a resolution of 0.3528 mm per pixel. The speed of sound images have a resolution of 1 mm per pixel.

4.3.1

Dataset 1

According to the Karmanos Cancer Institute, the breast of dataset 1 is dense. This leads to difficulties in recognizing tumors located below the high density tissue. For a non–expert it is not possible to recognize a tumor in both,the speed of sound and X–ray image (see figure 4.6). The X–ray mammogram has been reviewed by a radiologist and the woman related to dataset 1 has a tumor of the size 46x20x40 mm3 on the left side of the right breast. The resolution of the X–ray mammograms differ from the other dataset and is 0.4416 mm per pixel.

Figure 4.6: This figure shows the X–ray mammogram and the projection of the speed of sound image of dataset 1. A tumor is not clearly visible in either image.

4.3.2

Dataset 2

Dataset 2 shows the left breast of a breast cancer patient. In the X–ray mammogram, a tumor is clearly visible on the left side of the breast (right side of the image). In the projection of the corresponding speed of sound image, this tumor can be noticed as well (see figure 4.7). The tumor size is 35x17x35 mm3_.

Figure 4.7: This figure shows the X–ray mammogram and the projection of the speed of sound image of dataset 2. In the X–ray mammogram, a tumor of the size 35x17x35 mm3 is clearly visible on the left side of the breast. In the projection of speed of sound this tumor can be noticed as a slightly brighter area.

4.3.3

Dataset 3

In the X–ray mammogram of dataset 3, two tumors are visible. In the projection of the speed of sound image, the bigger tumor with a size of 20x21x24 mm3is clearly noticeable whereas the tumor which is located closer to the chest wall can barely be seen (see figure 4.8).

distance between the single slices is 1.2 mm instead of 1 mm as in the case for dataset 1, 2 and 4. This is due to the size of the breast and the attempt to have the same number of slices for every three–dimensional speed of sound image.

Figure 4.8: The X–ray mammogram of dataset 3 clearly renders two tumors. However, in the projection of the speed of sound image only the tumor located closer to the nipple is clearly visible; the second tumor is harder to detect.

4.3.4

Dataset 4

The right breast of dataset 4 has a large volume of glandular tissue which complicates the detection of a tumor (see figure 4.9). According to the radi- ologist, there is a tumor on the left side of the breast with a size of 19x15x22 mm3_.