| This is a slice of the raw CT data, after being restricted to the vicinity of the colon. | |
| This is the result of thresholding the slice to make the wall-to-interior boundary sharper. | |
| This is the result of applying the Sobel edge transform to the thresholded slice to identify the boundary between the colon interior and the colon wall. The Hough transform can now be applied to identify circular features in this boundary, such as polyps. | |
| This is the result of thresholding the Sobel output; having lots of completely zero data makes the Hough transform go faster. | |
| This is the same thresholded Sobel output, zoomed in on the polyp. The polyp is the bump on the horizontal boundary, while the sharp salient on the left is a normal haustral fold. | |
| This is the result of applying the Hough transform to the thresholded Sobel output, where we're looking at radius-2-pixel circles and I've drawn a red circle around the peak indicating the polyp. It's not immediately obvious how the map will be able to easily distinguish between the peak for the polyp and the bright areas corresponding to the haustral fold, but the fact that the polyp has only one isolated pixel of intensity turns out to be useful. | |
| This is another part of the Hough transform output, this time looking at radius-3-pixel circles. Notice that the peak for the polyp is more diffuse, so the real radius is more like 2 pixels, as in the above image. | |
| This is the result of applying the Laplace transform to the radius-2 Hough output. Notice that the polyp's peak is now higher than most of the fold, because while it was surrounded by lower intensities, the fold had a continuous region of high intensity. This image is ready for finding absolute peaks (since there are only a few as high as the actual polyp) and combining that information with the actual Hough-transform intensity to isolate circular formations, which we can then try to examine in the original CT data to find actual polyps. | |