Fourier neural operators to reconstruct tumor perfusion from 4D ultrasound data

Perfusion imaging is becoming increasingly important for monitoring how tumors respond to therapy. It provides information about the tumor’s blood flow and vascular state. Small changes in the blood vessels happen before any anatomical changes, e.g. changes in tumor size. One approach to perfusion imaging is three-dimensional dynamic-contrast-enhanced (DCE) ultrasound. Ultrasound is widely available, low-cost and safe to use at the bedside. In 3D DCE ultrasound, a contrast agent is injected and its flow through the tumor is recorded over time. This results in 4D data consisting of the three spatial dimensions and the time evolution. This thesis investigates Fourier Neural Operators as surrogate models for partial differential equation models to solve the inverse problem of estimating tissue-specific parameters, such as blood flow, from these measurements.

Subjects

Inverse Problems

3D dynamic contrast enhanced ultrasound

neural operators

perfusion imaging

DDC Class

610: Medicine, Health

006.3: Artificial Intelligence

Lizenz

https://creativecommons.org/licenses/by/4.0/

Name

Masterarbeit_JudithDeimel.pdf

Size

22.82 MB

Format

Adobe PDF

Options

Fourier neural operators to reconstruct tumor perfusion from 4D ultrasound data