Abstract |
In this thesis, we utilize existing tools to simulate a data acquisition process and
extend their capabilities to a virtual phantom with human anatomy characteristics.
Our proposed method involves the creation of a computational male pelvic phantom
and the utilization of advanced software to simulate magnetic resonance imaging
(MRI) processes, by solving Bloch equations. This work has the potential for important
future applications regarding the address of standardization challenges in
radiomics, contributing to explainability, robustness, and repeatability in the extraction
of radiomic features.
To accurately represent pelvis anatomy, we employed clinical MRI images, leveraging
the benefits of tomography and advanced medical image processing techniques.
Specifically, we employed segmentation and labeling methods to delineate and differentiate
each tissue, faithfully reproducing the spatial distribution of muscles and
fat—the primary components constituting the pelvis. In MRI, each tissue exhibits
distinct parameters such as T1, T2, T2*, and proton density, which are related to the
signal and the intrinsic contrast of each tissue. In this study, we assigned tissue properties
that relied on established values from the literature. For simulation purposes,
we utilized the specialized MRI simulator software, MRiLab. The implementation of
the fundamental clinical sequence, Spin Echo (SE), coupled with the investigation of
three distinct contrasts (T1-weighted, T2-weighted, and proton density-weighted),
forms a well-examined study. In our experimental results, we showcase the MRI
images generated through the simulation of the male pelvic phantom and highlight
specific applications in image analysis.
The methodology employed for phantom creation can be extended for deployment
in any body region, allowing the digital reproduction of patient-specific or requested
anatomical structures for various research studies. Leveraging the MRI simulator
facilitates the application of clinical MRI protocols and the generation of necessary
image data-sets. The potential applications and versatility of our phantom are extensive,
spanning from computational measurements and physics estimations to the
generation of synthetic data for training artificial intelligence models. Furthermore,
its utility extends to preclinical trials and educational purposes. In conclusion, our
phantom could be valuable in radiomic analysis. Its application emerges as a potential
solution to standardization challenges in radiomic features and facilitates a
comprehensive examination of variability issues across various clinical protocols.
|