Prostate magnetic resonance imaging analysis using deep learning

Prostate cancer (PCa) is one of the most common non-skin cancers in men, presenting a global healthcare challenge. Machine learning based prostate magnetic resonance imaging (MRI) analysis includes segmentation, PCa detection, grading, and recoherence prediction. The project focuses on prostate glan...

Full description

Saved in:
Bibliographic Details
Main Author: Li, Huanye
Other Authors: Huang Weimin
Format: Final Year Project
Language:English
Published: Nanyang Technological University 2022
Subjects:
Online Access:https://hdl.handle.net/10356/158297
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
Description
Summary:Prostate cancer (PCa) is one of the most common non-skin cancers in men, presenting a global healthcare challenge. Machine learning based prostate magnetic resonance imaging (MRI) analysis includes segmentation, PCa detection, grading, and recoherence prediction. The project focuses on prostate gland segmentation, prostate lesion segmentation, and prostate lesion classification. Image segmentation is a process of determining regions or boundaries of the area of interest, which is the first step in many clinical decision systems. Prostate lesion detection performs disease detection on given MRIs to assess the probability of certain disease. In this project, we used Transformer UNet as backbone to develop two models, ResUNet and TransUNet for the three tasks. Our experiment shows that the combination of Transformer and UNet yields improved segmentation performance as compared to using pure UNet, while ResNet based CNN outperforms traditional CNN for both segmentation and classification tasks. With the employment of loss function combination, neighboring slices input and hard case augmentation, the proposed algorithm achieved comparable performance as state-of-arts. Chapter 2 "literature review" of this report about machine learning in prostate MRI has been expanded and published in the journal of Diagnostics, by Multidisciplinary Digital Publishing Institute (MDPI), with the paper title “Machine Learning in Prostate MRI for Prostate Cancer: Current Status and Future Opportunities”, where I am the first author.