Available at: https://digitalcommons.calpoly.edu/theses/3379
Date of Award
6-2026
Degree Name
MS in Electrical Engineering
Department/Program
Electrical Engineering
College
College of Engineering
Advisor
Xiao-Hua (Helen) Yu
Advisor Department
Electrical Engineering
Advisor College
College of Engineering
Abstract
Image registration refers to the process of aligning multiple images of the same scene acquired at different times, from alternative viewpoints, or by various sensors. In the medical field, image registration is often employed to track the progress of diseases such as tumor growth by aligning multi-temporal image scans. Due to the impact registration accuracy has on clinical decisions and diagnosis, developing robust medical image registration methods remains an active and important area of research.
This thesis investigates the application of artificial neural networks for MRI (Magnetic Resonance Imaging) image registration through affine transform parameter estimation. Three distinct neural network architectures are considered, including a Multilayer Perceptron (MLP), a Convolutional Neural Network (CNN), and a Residual Network (ResNet). A modified CNN model trained on a mixed noise distribution is proposed, which yields the best performance across all evaluated models on clean MRI images, achieving a 16.9% reduction on total loss function for clean imagery, while also outperforming all evaluated architectures with respect to noise robustness, regardless of the noise type or severities investigated
Included in
Artificial Intelligence and Robotics Commons, Biomedical Commons, Signal Processing Commons