Smart algorithms clean up medical images
Work now underway by Dr. Michael Vannier at the Mallinckrodt Institute of Radiology (St. Louis, MO) aims to enhance computed-tomography (CT) images by significantly reducing image streaking. Image streaks are caused by attenuation of x-rays emanating from a scanner. When an x-ray beam is attenuated strongly, few photons reach the detectors, resulting in a loss of information. When the resultant image is computed, this loss of data appears as dark and bright streaks, edge gradients, or image artifacts.
To overcome these limitations, metal-artifact-reduction (MAR) techniques are used. MAR reduces the artifacts created by metal implants in a human body, such as braces and dental restorations, by extending the range of gray scales used for image display. "Streak artifacts," says Vannier, "are removed by interpolating raw data in the shadows of the metal object with adjacent raw data that do not contain the source of the artifact." The "removed" metal object is then scaled down in CT density and added back to the image.
"Among the methods now being used to perform MAR are polynomial interpolations," says Vannier. However, such iterative methods are susceptible to noise. Vannier favors a corrective approach in which dental implants in projection data are modeled using previously known data that can be applied to the CT data. These previously known data would include the shape and x-ray linear absorption coefficients of the dental implant, for example. The dental object is modeled using global pattern theory and CT images reconstructed by direct filtered back-projection so the implant characteristics can be determined preoperatively.