Computer-aided diagnosis of prostate cancer via machine learning using multiparametric MRI

Muftah, Asmail 2023. Computer-aided diagnosis of prostate cancer via machine learning using multiparametric MRI. PhD Thesis, Cardiff University. Item availability restricted.

	PDF - Accepted Post-Print Version Restricted to Repository staff only until 26 January 2025 due to copyright restrictions. Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (3MB)
	PDF (Cardiff University Electronic Publication Form) - Supplemental Material Restricted to Repository staff only Download (252kB)

Abstract

Prostate cancer (PCa) is a significant cause of male cancer death. Magnetic Resonance Imaging (MRI) can enhance the diagnosis of significant PCa, reducing over-diagnosis and biopsy risks. The manual analysis of MRI data (T2W, ADC map, and High B Value images), however, increases radiologists’ workload and requires substantial training. This study aims to develop automated techniques for classifying and segmenting PCa using multimodal MRI images, assisting radiologists in diagnosing PCa and formulating treatment plans. For classification of PCa, we employ two approaches: traditional machine learning with handcrafted features, and deep learning based on convolutional neural networks (CNNs), pre-trained or trained from scratch. This work investigates the utility of mp-MRI modalities, in the diagnosis of prostate cancer. To evaluate the potential advantages of utilising multiple MRI imaging modalities, our study examines the performance of each MRI modality (T2W, ADC and High B Value) individually and in combination. With regard to segmentation, we develop and evaluate PCa segmentation method using mp-MRI modalities. Deep learning techniques, specifically three different UNet architectures, are employed to address the challenges of PCa segmentation with MRI. To improve the segmentation performance, we modify all three UNet architectures by reducing the number of their parameters, which makes them lightweight models. Then we add expert-annotated prostate gland segmentation as a feature for the UNet models. Additionally, a new evaluation method was introduced in this study - a novel per-region detection evaluation metric that will complement per-pixel evaluation metrics and provide a more comprehensive and practically useful evaluation of machine learning models for PCa segmentation. The results demonstrate outstanding performance in the classification of early stage PCa and clinically significant (cs-PCa) cases, with Area under the ROC Curve (AUC) values exceeding 0.99 for both types of cancer. It is important to note, however, that these results are based on a relatively small dataset, which is a limitation of this study. In cs-PCa segmentation, the proposed methods achieved a DICE Coefficient of 0.569. These findings highlight the promising potential of the developed techniques for PCa classification, given the very high results, even if limited to a small dataset, and also for PCa segmentation. Further work to extend these results to a wider range of cases or larger datasets and validate them fully for clinical practice is necessary.

Item Type:	Thesis (PhD)
Date Type:	Completion
Status:	Unpublished
Schools:	Computer Science & Informatics
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Date of First Compliant Deposit:	26 January 2024
Last Modified:	30 Jan 2024 10:11
URI:	https://orca.cardiff.ac.uk/id/eprint/165874

Actions (repository staff only)

Edit Item