Fig. 3. Radon transform analysis. (A) Pixel grayscale values are summed along rows (in the direction of the green arrows), creating a column of row-intensity values. This is done iteratively as the image is rotated through 179°. (B) The resulting set of row-intensity columns is arrayed by rotation angle (orientation), and these columns are summed to create a density distribution (C). The peak of the density distribution therefore corresponds to the primary orientation of objects within the image. (D-G) This method reliably detects the orientation of radially arrayed struts (D), regardless of induced noise (E) or the removal of the image center (F), but will not produce a false signal for pure noise (G).