Avaliação de desempenho na reconstrução de imagens 2D de tomografia computadorizada utilizando programação massivamente paralela CUDA
The present study presents an analysis and evaluation of the performance in the reconstruction of 2D images of computed tomography, using a massively parallel approach in a GPU applying CUDA technology, comparing with a parallel and sequential approach in a conventional CPU. The quality of the gener...
| Autor principal: | Cordeiro, Alexssandro Ferreira |
|---|---|
| Formato: | Trabalho de Conclusão de Curso (Graduação) |
| Idioma: | Português |
| Publicado em: |
Universidade Tecnológica Federal do Paraná
2020
|
| Assuntos: | |
| Acesso em linha: |
http://repositorio.utfpr.edu.br/jspui/handle/1/12478 |
| Tags: |
Adicionar Tag
Sem tags, seja o primeiro a adicionar uma tag!
|
| Resumo: |
The present study presents an analysis and evaluation of the performance in the reconstruction of 2D images of computed tomography, using a massively parallel approach in a GPU applying CUDA technology, comparing with a parallel and sequential approach in a conventional CPU. The quality of the generated images was evaluated using the Peak Noise Ratio (PSNR) evaluation metrics, method used to verify the pixel differences through the mean square error (MSE); And by the structural similarity index (SSIM) method, used to verify the similarity of the images from loss of luminance, correlation, distortion and contrast distortion. We also analyzed the reconstructions with the Float 32 bits and Double 64 bits data types to validate the performance and quality of the generated images with the increase of the decimal places. Reconstructions of 2D computed tomography images were performed using the Filtered Back Projection (FBP) algorithm, using the Radon transform and convolution filter to remove image noise. The results indicate that the GPU in the Float 32 bits data type has the best performance among the approaches used. In Double 64 bits the parallel CPU approach performed better in comparison to GPU, but the GPU remained close to parallel CPU approach. |
|---|