www.MathResolutions.com Software Products for the Radiological Sciences

Search

Example of Image Fusion

CT to MRI

Shown below are the steps in achieving image fusion between a CT scan series of transverse scans, the skin surface shown on the left below, and an MRI series of coronal scans whose skin surface is also shown below on the right.

The CT series also had a stereotactic frame attached. Because the MRI series did not have a stereotactic frame, we are going to here match the skull and eye surfaces to achieve image fusion. We use the isosurface feature to produce a CT skull. But we do not want to pick up the stereotactic frame in the surface. To eliminate the stereotactic frame we first copied the skin surface and added a 0.5 cm margin to it. But we did have to first edit the skin surface contours to eliminate the points where the stereotactic frame attached to the patient's skull. Using the body surface with margin, we generated an isosurface from the image set with a restriction of the isosurface to inside the body surface with margin. We needed to add the margin to the body surface as the skull gets close to the skin in places and we would otherwise have omitted some skull surface. The skull surface is triangulated and a triangle reduction algorithm is used to reduce the number of triangles. The left and right eyes are outlined. Shown is the resultant skull and eye surfaces.

Next we need a skull surface from the MRI images. To do this we reverse the contrast of a MRI slice and adjust the contrast so that the skull appears white as it does on the CT scans. We use an automatic outlining routine to pick up the skull outline and automatically repeat the process on all the MRI scans. The contours were also edited to eliminate unwanted features. Shown is an MRI slice with a skull outline.

Shown is the skull surface resulting from the MRI skull contours. It is important that one of the surfaces to be matched is fairly clean of unrelated structures. We have used the CT skull in this case. For each matched surface, one surface, the cleaner and more complete surface, is designated as the template surface. For the corresponding surface in the other image series, points from the surface are taken. The distance from these points to the matching surface is used in a down hill search method to find the best transformation between two image series.

The two skull surfaces can be viewed together and manipulated manually with screen controls to the same approximate position. The eye contours make this manual positioning easier when rotating for two different orthogonal views. Then a down hill search method is used that considers the list of matched surfaces, here matched skull surfaces and matched eye contours, to find the best correlation between the two image sets. Shown are the two skull surfaces after the solution was found. We do note that there appears to be some difference in the shape of the CT and MRI skulls. We believe that this could be due to some distortion in the MRI scans and possibly to the different thickness of skull that CT and MRI will produce from their respective image data.

From the image fusion solution we can then reformat a plane from the CT scan image set in the MRI series for the same corresponding plane, shown below with the CT scan on the left, the reformatted MRI scan on the right.

Next we overlay these two images on top of each other in a checkerboard pattern, so that every other square shows the same image data, with adjacent squares alternating showing CT and MRI data from the two images.

Return to Prior Page
Return to Home Page