GUIMARÃES, Daniel Marques Vasconcelos; http://lattes.cnpq.br/4103540054940166; GUIMARÃES, Daniel Marques Vasconcelos.
Resumo:
The presence of metallic artifacts (noise) in computed tomography disturbs the medical
diagnosis. When an image is reconstructed using incomplete data, unnatural changes in
appearance occur, known as artifacts, which are usually seen as light or dark bands in the
image. Errors in CT scans affect the diagnostic capacity and make it difficult to accurately
distinguish the types of tissues, impairing radiation planning for the treatment of cancer,
or orthopedics or dentistry exams, due to the high quality required of the CT scan image.
Several methods for reducing metallic artifacts in medical images have been developed in
the last 40 years. This dissertation proposes a method of three-dimensional interpolation
based on registration between adjacent slices for reconstruction of computed tomographies. Metallic artifacts were created in tomography without noise, affecting 1, 2, 3, 4, 5 and 9 consecutive slices of CT scans, for later reconstruction and comparison with the classical linear interpolation method, along with other methods of information retrieval through the Radon transform and the Fanbeam projection in the reconstruction of slices of computed tomography affected by metallic artifacts. Quantitative analyzes using the correlation, PSNR and absolute mean error metrics; and qualitative analyzes were performed to evaluate the performance of the record-based interpolation method. The results show that the proposed method presents better results with the metrics mentioned above in relation to linear interpolation from the reconstruction of at least 3 consecutive slices of computed tomography. With the use of interpolation methods, the reconstructed computed tomographies showed the effect of aliasing, this effect being more attenuated in record-based interpolation than in linear interpolation.