Synthetic Correlated Diffusion Imaging for Prostate Cancer Detection and Risk Assessment
| dc.contributor.author | Gunraj, Hayden | |
| dc.date.accessioned | 2023-08-31T15:40:54Z | |
| dc.date.available | 2023-08-31T15:40:54Z | |
| dc.date.issued | 2023-08-31 | |
| dc.date.submitted | 2023-08-17 | |
| dc.description.abstract | Prostate cancer (PCa) is the second most common form of cancer among men worldwide and the most frequently diagnosed cancer among men in 112 countries. While the overall 5-year survival rate for prostate cancer is very high, prognosis varies considerably depending on how early PCa is diagnosed and how aggressively it develops. As such, PCa screening is critical for early detection and treatment of PCa. However, many PCas develop slowly and pose a minimal risk of PCa-related mortality, in which case treatment can be limited to active surveillance of tumour development. Over the last few decades, magnetic resonance imaging (MRI) been used extensively for PCa screening and assessment. In particular, multi-parametric magnetic resonance imaging (mpMRI), where multiple MRI modalities are acquired, is commonly used for PCa imaging. However, the use of mpMRI requires radiologists to interpret multiple MRI images in parallel, resulting in increased inter-observer variability. This is especially true for radiologists with less experience interpreting prostate MRI images. In an effort to address these concerns, a computational MRI modality known as correlated diffusion imaging (CDI) was introduced, with initial results showing promise for CDI as a PCa screening tool. However, CDI is uncalibrated and strongly dependent on the underlying MRI protocols used to compute it, which leads to inconsistencies across different protocols and significant inter- and intra-patient variability. In this thesis, a computational MRI technique known as synthetic correlated diffusion imaging (CDIs) is introduced. CDIs extends CDI through the addition of synthetic DWI and per-patient calibration, thereby providing flexibility and consistency beyond that of CDI. Additionally, a gradient-based optimization framework is developed through which the parameters of CDIs may be optimized for downstream clinical tasks. The proposed CDIs and optimization framework were evaluated against current standard MRI modalities using a clinical MRI dataset comprising 200 PCa patients. Through clinical interpretation by an experienced radiologist, CDIs was found to provide better tissue contrast between healthy, low-risk PCa, and high-risk PCa than standard MRI modalities. This suggests that CDIs provides visual indications of PCa and PCa risk level, which may allow radiologists to make more accurate and consistent conclusions from imaging alone. CDIs may also be used to guide prostate biopsies, potentially indicating better biopsy locations and reducing the number of biopsies required. Upon quantitative evaluation, CDIs achieved a voxel-level area under the receiver operating characteristic curve (AUC) of 0.8446 for separation of healthy and PCa tissue, representing an increase of 0.0315 (p<0.0001) over the best-performing standard MRI modality and indicating the potential of CDIs for PCa screening and diagnosis. Moreover, CDIs achieved a voxel-level AUC of 0.8530 for distinguishing between high- and low-risk cancers, representing an increase of 0.1590 (p<0.0001) over the best-performing standard MRI modality and indicating the potential of CDIs for PCa risk assessment. These results suggest that CDIs may improve voxel-level identification of PCa, which is valuable for PCa localization and segmentation. Moreover, machine learning models trained on CDIs images can benefit from this improved voxel-level contrast, potentially achieving better diagnostic, prognostic, or segmentation performance than models trained on standard MRI images. | en |
| dc.identifier.uri | http://hdl.handle.net/10012/19816 | |
| dc.language.iso | en | en |
| dc.pending | false | |
| dc.publisher | University of Waterloo | en |
| dc.relation.uri | https://www.kaggle.com/datasets/hgunraj/cancer-net-pca-data | en |
| dc.relation.uri | https://wiki.cancerimagingarchive.net/pages/viewpage.action?pageId=23691656 | en |
| dc.subject | magnetic resonance imaging | en |
| dc.subject | diffusion-weighted imaging | en |
| dc.subject | prostate cancer | en |
| dc.subject | machine learning | en |
| dc.subject | computational imaging | en |
| dc.title | Synthetic Correlated Diffusion Imaging for Prostate Cancer Detection and Risk Assessment | en |
| dc.type | Master Thesis | en |
| uws-etd.degree | Master of Applied Science | en |
| uws-etd.degree.department | Systems Design Engineering | en |
| uws-etd.degree.discipline | System Design Engineering | en |
| uws-etd.degree.grantor | University of Waterloo | en |
| uws-etd.embargo.terms | 0 | en |
| uws.contributor.advisor | Wong, Alexander | |
| uws.contributor.affiliation1 | Faculty of Engineering | en |
| uws.peerReviewStatus | Unreviewed | en |
| uws.published.city | Waterloo | en |
| uws.published.country | Canada | en |
| uws.published.province | Ontario | en |
| uws.scholarLevel | Graduate | en |
| uws.typeOfResource | Text | en |