Image Reconstruction

Field of View

CT images are reconstructed from approximately 1000 projections that are acquired as the x-ray tube rotates through 360º around the object (patient). The acquisition geometry is defined by the acquisition Field of View which is determined by the fan beam angle, and will determine the maximum possible size of reconstructed image. The acquisition FOV is typically 250 mm for head CT scans, but can be as large as 500 mm for body imaging. Increasing the acquisition FOV requires the use of a larger fan beam angle, achieved by adjustment of collimation, and increasing the number of detectors used when acquiring the raw projection data.

Once the projection data have been acquired, operators can choose the size of the reconstruction field of view (FOV). The majority of CT images have a reconstructed FOV that is set equal to the size of the patient being imaged, as depicted by the two left images in Figure G. In this example, the FOV of 250 mm together with a matrix size of 512 results in a pixel size of ~0.5 mm that corresponds to the achievable spatial resolution performance. If the acquired projection data set are used to reconstruct an image with a smaller field of view (right images in Figure G), the pixel size is now 0.25 mm (125 mm FOV divided by 512 matrix size), and the achievable resolution has improved by about a factor of two or so.

It is important to note that reconstructing with a smaller FOV must be done from the acquired (raw) projection data set, and not from the reconstructed images shown in Figure G (right), and can only be applied to the central region of the reconstructed image. There is also a limit to the improvement in spatial resolution that is achievable by reducing the reconstruction FOV, since additional factors (i.e., focal spot size, motion, and detector size) limit the achievable spatial resolution. As a rough guide, the achievable spatial resolution performance in normal CT imaging is ~ 0.7 lp/mm, and this can be approximately doubled to ~ 1.5 lp/mm by the use of the zoom feature. A corollary of this finding is that if the reconstructed FOV is selected so that it is larger than the object being imaged (Figure H), there will be a loss of spatial resolution performance being of the increased pixel size. In Figure H, increasing the reconstruction FOV from 350 mm (right) to 500 mm (left) increases the pixel size from ~0.7 mm to ~1 mm.

Figure G. Images of a skull phantom reconstructed with a FOV that encompasses the whole head (i.e., 25 cm), as shown by the two sections on the left, or zoomed into a smaller FOV (i.e., 12.5 cm) as shown by the two sections on the right.

 

Figure H. Images of a body phantom acquired on a four slice CT scanner with a 50 cm field of view. The two images on the left were reconstructed using a 500 mm reconstruction field of view, whereas as the images on the right used a reconstruction FOV of 350 mm.

Reconstruction Filter

CT makes use of filtered back projection reconstruction techniques, whereby each projection is convolved with a “filter”, and then back projected. When this procedure is performed for all 1000 or so projections, it is possible to achieve a perfect reconstruction of the scanned object. A special filter, known as the ramp filter, will achieve a perfect reconstruction, but will also contain high levels of image noise. It is possible to modify the shape of the reconstruction filter to reduce the amount of image noise, but at the price of some loss in spatial resolution performance. Most commercial CT scanners offer the user a choice of reconstruction filters where each filter offers varying degrees of image noise, but at the price of some loss of spatial resolution performance.

Figure I shows an example of CT images of a skull phantom, where one set of projection images was acquired and subsequently used to generate four sets of images using four reconstruction filters that are available on this commercial CT scanner. For each of the images shown in Figure I, the mean HU value in the region of interest remains constant with reconstruction filter (i.e., ~133 HU). Image noise, however, varies markedly with choice of reconstruction filter: Soft ~ 7.4 HU; Detail ~12 HU; Bone ~ 30 HU; and Edge ~ 55 HU. The amount of noise is thus seen to vary by more than a factor of seven, depending on the choice of reconstruction filter used to generate these filtered back projection images. Figure J and K show blow up images of bony structures that clearly illustrate the trade off between noise and spatial resolution performance offered by these reconstruction filters. Note the loss of sharpness when using a Soft reconstruction filter, and the noise images when using an Edge reconstruction Filter.

Figure L shows a series of CT images obtained using a body phantom. As with the skull phantom images, for each of the images shown in Figure L, the mean HU value in the region of interest remains constant with reconstruction filter (i.e., ~144 HU). Image noise, however, varies markedly with choice of reconstruction filter: Soft ~ 8.5 HU; Detail ~11 HU; Bone ~ 25 HU; and Edge ~ 39 HU. Figure M and N show blow up images of bony structures that clearly illustrate the trade off between noise and spatial resolution performance offered by the reconstruction filters on this commercial CT scanner.

Figure I. CT images of an anthropomorphic head phantom reconstructed from the same projection data set using four different types of reconstruction filter (Soft, upper left; Detail, upper right; Bone, lower left; Edge, lower right) on a GE LightSpeed CT scanner.

 

Figure J. Small region of interest of the skull phantom reconstructed using four filters (Soft, upper left; Detail, upper right; Bone, lower left; Edge, lower right)showing the trade off between good resolution (i.e. Edge) and low noise (i.e., Soft).

 

Figure K. Small region of interest of the skull phantom reconstructed using four filters (Soft, upper left; Detail, upper right; Bone, lower left; Edge, lower right)showing the trade off between good resolution (i.e. Edge) and low noise (i.e., Soft).

 

Figure L. CT images of an anthropomorphic body phantom reconstructed from the same projection data set of the spine using four different types of reconstruction filter (Soft, upper left; Detail, upper right; Bone, lower left; Edge, lower right) on a GE LightSpeed CT scanner.

 

Figure M. Small region of interest of the spine phantom reconstructed using four filters (Soft, upper left; Detail, upper right; Bone, lower left; Edge, lower right) on a GE LightSpeed CT scanner.

 

Figure N. Small region of interest of the pelvis phantom reconstructed using four filters (Soft, upper left; Detail, upper right; Bone, lower left; Edge, lower right) on a GE LightSpeed CT scanner.

rsna