Simulation framework of contrast enhanced ultrasound(CEUS) and microvascular flow for ultrasound localization microscopy (ULM) and deep learning
File(s)
Author(s)
Lerendegui, Marcelo
Type
Thesis
Abstract
Standard ultrasound imaging is a widely adopted noninvasive and safe modality, known for
its affordability, real-time capabilities, and ease of use. Despite these advantages, it faces
limitations such as low resolution due to the diffraction limit and a trade-off between frequency
and imaging depth.
Recent decades have witnessed notable advancements in ultrasound vascular imaging, in-
cluding Microbubble (MB) contrast agents that highlight vasculature śknown as Contrast En-
hanced Ultrasound (CEUS)ś and unfocused transmission techniques that increase framerate.
Inspired by optical imaging techniques, Ultrasound Localization Microscopy (ULM) emerged as
a novel approach capable of overcoming the diffraction limit, providing unprecedented in-vivo
resolution for microvascular ŕow.
While ULM has rapidly expanded, signiőcant challenges for clinical translation exist, such
as low acquisition rate and the need for algorithms capable of isolating, localizing, and tracking
MBs accurately. Many algorithms addressing these challenges were developed, but their per-
formance has not been consistently evaluated. Previous evaluation attempts have limitations
including datasets that neglect MB oscillation physics or focus solely on localization.
This work addresses these challenges by presenting tools for ULM development:
• An optimized simulation method for linear acoustic őelds.
• A comprehensive framework for realistic vascular structure generation, microvascular ŕow
simulation, and performance evaluation metrics.
• Analysis of inter-bubble distance effects on localization algorithms.
• An ULM localization and tracking challenge with realistic ultrasound data and bench-
marking methodologies.
• Three deep learning models for ULM localization, including a modiőed UNet, YOLOv3,
and a network predicting bubble locations on RadioFrequency (RF) data.
This work contributes valuable resources for overcoming limitations in ULM research, offer-
ing a realistic imaging simulator and comprehensive evaluation tools to advance algorithms for
future clinical translation, enabling researchers to create controlled experiments for systematic
exploration of algorithm performance.
its affordability, real-time capabilities, and ease of use. Despite these advantages, it faces
limitations such as low resolution due to the diffraction limit and a trade-off between frequency
and imaging depth.
Recent decades have witnessed notable advancements in ultrasound vascular imaging, in-
cluding Microbubble (MB) contrast agents that highlight vasculature śknown as Contrast En-
hanced Ultrasound (CEUS)ś and unfocused transmission techniques that increase framerate.
Inspired by optical imaging techniques, Ultrasound Localization Microscopy (ULM) emerged as
a novel approach capable of overcoming the diffraction limit, providing unprecedented in-vivo
resolution for microvascular ŕow.
While ULM has rapidly expanded, signiőcant challenges for clinical translation exist, such
as low acquisition rate and the need for algorithms capable of isolating, localizing, and tracking
MBs accurately. Many algorithms addressing these challenges were developed, but their per-
formance has not been consistently evaluated. Previous evaluation attempts have limitations
including datasets that neglect MB oscillation physics or focus solely on localization.
This work addresses these challenges by presenting tools for ULM development:
• An optimized simulation method for linear acoustic őelds.
• A comprehensive framework for realistic vascular structure generation, microvascular ŕow
simulation, and performance evaluation metrics.
• Analysis of inter-bubble distance effects on localization algorithms.
• An ULM localization and tracking challenge with realistic ultrasound data and bench-
marking methodologies.
• Three deep learning models for ULM localization, including a modiőed UNet, YOLOv3,
and a network predicting bubble locations on RadioFrequency (RF) data.
This work contributes valuable resources for overcoming limitations in ULM research, offer-
ing a realistic imaging simulator and comprehensive evaluation tools to advance algorithms for
future clinical translation, enabling researchers to create controlled experiments for systematic
exploration of algorithm performance.
Version
Open Access
Date Issued
2024-07-02
Date Awarded
2025-04-01
Copyright Statement
Attribution-NonCommercial-ShareAlike 4.0 International Licence (CC BY NC-SA)
Advisor
Tang, MengXing
Dunsby, Christopher
Sponsor
Engineering and Physical Sciences Research Council
Grant Number
Grant EP/T008970/1
Publisher Department
Department of Bioengineering
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)