University of Twente Student Theses
Iterative computed tomography reconstruction using deep learning
Moraru, Alexandru (2020) Iterative computed tomography reconstruction using deep learning.
|
PDF
1MB |
| Abstract: | In computed tomography it is important not only to obtain images of good quality but also to minimize the radiation dose given to the patient. Research efforts are dedicated to increase the quality of the reconstructed images and minimize the radiation exposure. This work addresses the problem of reducing the dose by using deep learning to correct the update term of the simultaneous iterative reconstruction technique (SIRT). The aim is to improve the output of the backprojection operator which does not rely on any prior knowledge about the object and distributes all the rays back into the volume under reconstruction uniformly. We propose a deep learning solution to correct the update term of the SIRT algorithm after the backprojection operator has been applied with the purpose to increase the image quality. We evaluate the quality of the images obtained with the proposed method using similarity measures between the low dose reconstructions obtained with the proposed method and the high dose reconstructions taken as ground truth. We also investigate whether the iterative scheme converges faster with the proposed modification. We obtained a structural similarity index (SSIM) of 0.725, a peak signal-to-noise ratio (PSNR) of 29.42 dB and a mean absolute error (MAE) of 92.69 HU which indicates that our method performs better than the classical SIRT algorithm. We also demonstrated that the proposed iterative scheme has the side benefit that it converges faster, achieving with three iterations the similarity that is obtained with the classical scheme with 115 iterations. |
| Item Type: | Essay (Master) |
| Clients: | UMC Utrecht, Utrecht, Netherlands |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Subject: | 54 computer science |
| Programme: | Electrical Engineering MSc (60353) |
| Link to this item: | https://purl.utwente.nl/essays/85407 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page