Denoising Low-Dose CT Images using Multi-frame techniques
| dc.contributor.advisor | Wahid, Khan | |
| dc.contributor.advisor | Babyn, Paul | |
| dc.contributor.committeeMember | ERAMIAN, MARK | |
| dc.contributor.committeeMember | Sarty, Gordon | |
| dc.contributor.committeeMember | Chen, Daniel | |
| dc.creator | Hasan, Ahmed Monjurul | |
| dc.date.accessioned | 2022-05-20T21:32:32Z | |
| dc.date.available | 2022-05-20T21:32:32Z | |
| dc.date.created | 2022-04 | |
| dc.date.issued | 2022-05-20 | |
| dc.date.submitted | April 2022 | |
| dc.date.updated | 2022-05-20T21:32:32Z | |
| dc.description.abstract | This study examines potential methods of achieving a reduction in X-ray radiation dose of Computer Tomography (CT) using multi-frame low-dose CT images. Even though a single-frame low-dose CT image is not very diagnostically useful due to excessive noise, we have found that by using multi-frame low-dose CT images we can denoise these low-dose CT images quite significantly at lower radiation dose. We have proposed two approaches leveraging these multi-frame low-dose CT denoising techniques. In our first method, we proposed a blind source separation (BSS) based CT image method using a multiframe low-dose image sequence. By using BSS technique, we estimated the independent image component and noise components from the image sequences. The extracted image component then is further donoised using a nonlocal groupwise denoiser named BM3D that used the mean standard deviation of the noise components. We have also proposed an extension of this method using a window splitting technique. In our second method, we leveraged the power of deep learning to introduce a collaborative technique to train multiple Noise2Noise generators simultaneously and learn the image representation from LDCT images. We presented three models using this Collaborative Network (CN) principle employing two generators (CN2G), three generators (CN3G), and hybrid three generators (HCN3G) consisting of BSS denoiser with one of the CN generators. The CN3G model showed better performance than the CN2G model in terms of denoised image quality at the expense of an additional LDCT image. The HCN3G model took the advantages of both these models by managing to train three collaborative generators using only two LDCT images by leveraging our first proposed method using blind source separation (BSS) and block matching 3-D (BM3D) filter. By using these multi-frame techniques, we can reduce the radiation dosage quite significantly without losing significant image details, especially for low-contrast areas. Amongst our all methods, the HCN3G model performs the best in terms of PSNR, SSIM, and material noise characteristics, while CN2G and CN3G perform better in terms of contrast difference. In HCN3G model, we have combined two of our methods in a single technique. In addition, we have introduced Collaborative Network (CN) and collaborative loss terms in the L2 losses calculation in our second method which is a significant contribution of this research study. | |
| dc.format.mimetype | application/pdf | |
| dc.identifier.uri | https://hdl.handle.net/10388/13974 | |
| dc.language.iso | en | |
| dc.subject | Multiframe, Low-Dose CT, | |
| dc.title | Denoising Low-Dose CT Images using Multi-frame techniques | |
| dc.type | Thesis | |
| dc.type.material | text | |
| thesis.degree.department | Biomedical Engineering | |
| thesis.degree.discipline | Biomedical Engineering | |
| thesis.degree.grantor | University of Saskatchewan | |
| thesis.degree.level | Masters | |
| thesis.degree.name | Master of Science (M.Sc.) |