MUST 2024

Jun 10, 2024, 12:00 PM → Jun 12, 2024, 5:00 PM Europe/Amsterdam

Eulerzaal (Centrum Wiskunde & Informatica)

The 4th International Workshop on Medical Ultrasound Tomography (MUST 2024)

Location: Amsterdam, NL

Dates: 10-12 June 2024

Type: In-person event

The forth edition of the International Workshop on Medical Ultrasound Tomography will be held in Amsterdam from 10th to 12th June 2024. The conference will be hosted by the Centrum Wiskunde & Informatica (CWI), the national research institute for mathematics and computer science in the Netherlands

MUST brings together research groups from around the globe that are engaged in the development of ultrasound tomography (UST) – an emerging medical imaging technique with immense clinical potential. The workshop is designed to be interactive and intimate, allowing for discussion, exchange of new ideas and fostering collaboration between researchers working on all parts of the development chain of UST, from mathematical modeling over image reconstruction, processing and analysis to instrumentation, clinical applications and commercialization.

Invited Speakers

Yonina Eldar	Ben Cox	Christian Böhm	Richard Lopata

Sponsors

MUST receives NWO Scientific Meetings and Consultations Domain Science funding (project number 21379)

Monday, June 10

Registration and Lunch

Plenary Presentation: Yonina Eldar (online)

Break

Applications & Multimodal Extensions

1 Enhancing Tissue Quantification with Ultrasound-Informed Quantitative Photoacoustic Imaging

Quantifying tissue composition is crucial for accurate medical diagnoses. Photoacoustic imaging, which uses pulsed light to generate sound waves from tissue, emerges as a promising technique for tissue quantification. It offers deep imaging capabilities, optical contrast, and acoustic resolution. Its integration with conventional ultrasound imaging leverages structural tissue information alongside the functional insights provided by photoacoustic imaging, enhancing diagnostic potential. However, the challenge lies in the complex interplay of light and sound propagation within tissues, making quantitative analysis difficult.

We present two strategies named Ultrasound-informed Quantitative Photoacoustic Imaging, utilizing ultrasound images to improve photoacoustic imaging quantification. The first method employs ultrasound-derived tissue boundaries to model light propagation. We present a fluence compensated dual-wavelength oxygen saturation imaging method, utilizing structural information from the ultrasound image, and prior knowledge of the optical properties of the tissue with a Monte-Carlo based light propagation model. This approach has been tested on phantoms, in-vivo experiments on mouse thigh tissues under an oxygen challenge, and applications on human volunteers. The proposed method was found to improve the oxygen saturation imaging accuracy. The second strategy uses arterial blood as an internal reference, using known values of oxygen saturation and hematocrit to estimate the optical properties of blood. This internal marker aids in accurately determining the optical properties of unknown chromophores within tissues. We implemented this prior information in a gradient-based iterative optical inversion method with diffusion approximation for light propagation modelling. We demonstrate this method in the context of carotid plaque imaging using simulated phantoms and present our preliminary experimental results.

Our findings underscore the potential of integrating ultrasound imagery with photoacoustic imaging for advanced tissue quantification. This synergy not only improves the accuracy of existing quantitative photoacoustic imaging methods but also opens new avenues for medical applications such as plaque composition analysis and oxygen saturation imaging.

Speaker: Francis Kalloor Joseph (Erasmus MC)

2 Hybrid photoacoustic-ultrasound breast tomography

We report on a hybrid photoacoustic and ultrasound-transmission tomographic system (PAM3). The photoacoustic component has multi-wavelength imaging capability, and implements substantial technical advancements in important hardware and software sub-systems. The ultrasound component enables for the first time, a three-dimensional sound speed map of the breast to be imaged. This map is used in the photoacoustic reconstruction to correct for inhomogeneities. The results demonstrate deep photoacoustic breast imaging. We present a system overview and demonstrate the in vivo performance of the imager.

Speaker: Srirang Manohar

3 Ultrasound computed tomography applied to peripheral nerve imaging

Background:
Peripheral nerves play crucial roles in motor and sensory functions, and their clinical imaging typically involves ultrasound and MRI. Ultrasound has limitations due to shallow depth, poor spatial resolution, and operator-dependent acquisition, while MRI is constrained by imaging time and motion artifacts. Ultrasound Computed Tomography (USCT) presents a promising solution for achieving precise imaging of peripheral nerves, attributable to its high resolution, fast scanning speed, and volumetric imaging capabilities. This study delineates the utilization of USCT in limb scanning to acquire high-resolution peripheral nerve images.
Material and Method:
The in vivo experiment utilized a circular probe with 2048 elements and a 220mm diameter, operating at a central frequency of 3MHz with water as the coupling agent. The limb under examination was positioned at the center of the probe for spiral scanning. A thorough scan of the elbow joint and the knee joint was conducted. Subsequently, several visualization operations, such as segmentation and 3D reconstruction, are used to visualize the nerves.
Results:
The experimental results demonstrate that USCT can facilitate precise imaging of the principal nerves, allowing the distinction of structures such as the radial nerve, median nerve, ulnar nerve, and peroneal nerve. The single-layer image acquisition time was less than 1s. All scans were completed within the 90s. A certain degree of patient motion can only lead to adjacent layer image position mismatch, but not single-layer image damage.
Conclusion:
Peripheral nerves, typically delicate structures measuring less than 5mm, present a challenge in imaging due to patients' inability to keep their limbs still. So achieving high-resolution and high-speed imaging is crucial in nerve imaging. Experiments have shown that USCT can achieve fast imaging of peripheral nerves. Motion artifacts caused by patient shivering are avoided, and high-resolution peripheral nerve imaging is achieved, which is expected to help clinical neurological disease diagnosis.

Speaker: Hui Zhang (Huazhong university of Science and Technology, Department of Biomedical Engineering, collage of Life Science and Technology.)

4 Preliminary research on in vivo brain imaging using ultrasound computed tomography

Background:
Intracranial soft tissue imaging using traditional ultrasound is a difficult task due to the presence of the skull, which strongly reflects and attenuates the ultrasound signal. To address this problem, we have conducted preliminary research on full waveform inversion (FWI)-based ultrasound computed tomography (USCT) for in vivo human brain imaging.

Material and Method:
We have developed a prototype system for transmitting and receiving signals in single-channel transcranial ultrasound tomography. The prototype system consists of a pair of piston probes, an electric rotary stage, an optical tracking positioner, and a transmitting and receiving system. The probes operate at a center frequency of 500kHz. The rotary stage is used to rotate the transmitting and receiving probes. The optical tracking positioner records the position of the transducer relative to the human head, which is located in the center of the rotary stage. The transmitting and receiving system transmits and captures the signals through the brain. Using the received signal, a sound-speed image of the brain is reconstructed using FWI.

Results:
We conducted experiments on the head of one healthy volunteer. The transmitting and receiving probes were positioned along the radial direction. A total of 64 transmit events were configured, with each event consisting of 128 received traces. Although the preliminary result still has poor image quality, it shows the potential of in vivo brain imaging using ultrasound computed tomography.

Conclusion:
Traditional ultrasound has limitations in visualizing intracranial soft tissue. Full waveform inversion-based ultrasound computed tomography has the potential to enable in vivo imaging of the human brain, making it a promising non-invasive modality for clinical applications in brain diseases.

Speaker: Xiaolu Zeng (Huazhong University of Science and Technology, School of Life Science and Technology)

5 In Vivo Bistatic Multi-Aperture Ultrasound Imaging and Elastography of the Abdominal Aorta

Abdominal aortic aneurysms (AAA) are large dilatations of the abdominal aorta, that are typically asymptomatic until a life-threatening rupture occurs. Knowledge of AAA geometry and local mechanical wall parameters using ultrasound is paramount for risk stratification and intervention planning. However, such an assessment is limited by the lateral lumen-wall contrast and resolution of conventional ultrasound. Fundamentally, a semi-tomographic set-up using multiple apertures would produce images with improved angular coverage of the vessel wall. In this study, this concept is assessed by introducing dual-aperture bistatic imaging, in which two curved array transducers alternately transmit and both probes receive simultaneously on each transmit event, which allows for the reconstruction of four ultrasound signals.

The performance of dual-aperture bistatic ultrasound imaging and elastography was assessed in vivo in 20 healthy volunteers and 40 patients with an AAA. Automatic probe localization was achieved by optimizing the coherence of the trans-probe data, using a gradient descent algorithm. To measure deformation and strain inside the aortic wall, motion tracking was performed on the four individual ultrasound signals, after which the respective axial displacements were compounded. To mitigate the impact of aberrations, an Eikonal-based beamforming method was tested in vivo.

The automatic probe localization approach improved the lumen-wall contrast of the trans-probe data by 3.7 dB, compared to manual registration. The results in healthy volunteers and AAA patients show both the feasibility and promise of multi-aperture ultrasound 1. The lumen-wall generalized contrast-to-noise ratio (gCNR) is increased by 40% on average, yielding more homogeneous and more accurate strain estimates (+12dB), compared to conventional ultrasound. Eikonal-based beamforming improved the alignment of image features in coherent dual-aperture image fusion, and increased lumen-wall contrast by 1.5 dB. Our future vision is the use of this concept while imaging the abdomen with a large, flexible transducer and using 3-D ultrasound

Speaker: Vera van Hal (Eindhoven University of Technology)

Poster Session, Lab Tours & Reception

6 Automatic tumor detection based on optimized YOLOv5 in breast ultrasound tomography

Breast cancer has surpassed lung cancer as the leading cause of global cancer incidence in 2020. Traditional handheld 2D ultrasound imaging is one of the major imaging tools for breast cancer screening in China. Compared with the traditional method, ultrasound tomography (UST) provides three-dimensional reproducible images of higher quality, simplifying and standardizing the image acquisition process. As a result, UST is a potential imaging technique for the early detection of breast cancer. This study aims for developing a deep learning method to automatically detect breast cancer from UST images, free radiologists from the incomprehensible three-dimensional data. Breast ultrasound tomography imaging system developed in HUST(Huazhong University of Science and Technology, Wuhan, China) with a scanning slice interval of 2 mm has scanned 40 patients with either benign or malignant breast tumor to get 3D data, each patient with about 30 2D slices with a size of 2048×2048. A dataset including 279 pairs of images and labels from 40 patients is established. An optimized YOLOv5-based method is proposed to detect breast tumors. In order to incorporate 3D context information efficiently, the target images containing tumors with their two neighbouring slices are grouped into one pseudo 3D image for the training. Manually labeling bounding boxes for the tumor on the center slice in each pseudo image are used as labels. The dataset is randomly divided into training (80% of patients) and test (20% of patients). The evalutation results on the testing dataset show that, the proposed method is able to yield detection recall, precision, and mAP (mean Average Precision) of 71.6%, 55.6%, and 57.1%, respectively, implying that the proposed method can potentially serve as a preliminary helper for doctors to automatically locate tumor on breast UCT images.

Speaker: Gaoxiang Xie

7 Multichannel end-to-end AI-based super-resolution

Super-resolution ultrasound imaging represents a significant advancement in the
field of medical imaging, particularly through the application of ultrasound localization
microscopy (ULM). However, ULM's requires sparse microbubble distributions, which in
turn, imposes prolonged acquisition times to adequately image the full
microvasculature. We introduce a novel super-resolution methodology that
overcomes traditional ULM limitations by employing a direct deconvolution technique
on all radiofrequency (RF) channels obtained by single plane-wave imaging. We
leverage a deep learning physics-based approach, which integrates Stolt's FK
migration beamforming algorithm into the network architecture. Focusing on
low-frequency ultrasound (1.7 MHz), our research targets deep imaging capabilities
(up to 10 cm) within a dense cloud of monodisperse microbubbles, accommodating
up to 1000 microbubbles within the 2D measurement volume. Our approach utilizes
a simulation framework that encompasses a broad spectrum of acoustic pressures
(5-250 kPa) to accurately model the complex, nonlinear response of resonant,
lipid-coated microbubbles. Previously, published work uses single-channel
deconvolution with a 1D dilated convolutional network. The core innovation of this
study lies in replacing this 1D network by various 2D architectures operating on all
RF lines simultaneously in order to exploit the correlation among transducer
elements and allowing end-to-end learning of microbubble location. We expect this
approach to reduce detection uncertainty and to more effectively handle high bubble
concentrations, leading to better localization in more realistic scenarios.

Speaker: Anass Hameddine (Physics of Fluids (POF), Faculty of Science and Technology (TNW), University of Twente.)

8 Simultaneous estimation of initial pressure and speed of sound in photoacoustic tomography

Photoacoustic tomography (PAT) is a biomedical imaging technique utilising the photoacoustic effect that can be induced by an external light pulse. The aim in PAT is to reconstruct an initial pressure distribution that has been generated in the target by illumination of light. This is carried out by measuring the propagating pressure waves on the boundary of the target, and performing a computational image reconstruction.

Conventionally in PAT, acoustic propagation is modelled in a homogeneous medium. This approximation, however, is not generally true in realistic targets. For example, assuming wrong speed of sound can cause artefacts in the reconstructions. One possible approach to mitigate these artefacts is to estimate the speed of sound in the target. However, estimation of the initial pressure and speed of sound distributions simultaneously is generally an unstable problem if no additional information is utilised. In this work, we propose a methodology for simultaneous estimation of initial pressure and speed of sound, when multiple illuminations of light are utilised. Results of numerical simulations are shown.

Speaker: Miika Suhonen (University of Eastern Finland)

9 Next KIT 3D USCT v3.2 Mechanical frame - Compatibility&Simplification

Motivation

The KIT 3D ultrasound computertomography project is kickstarting the build up of a second machine of the 3D USCT III device generation, called v3.2. Extensive experience with their successfully commissioned device v3.1 regarding mechanical and handling properties were aquired over the last month and years: integration, maintainance, updates, transport tasks, many test measurements and a started clinical trial were among the device's life-cycle interactions. This broad range of interaction activities with the complex medical device lead to many experiences and insights regarding engineering and practicability aspects.

Methods

These experience should now flow back as improvements for the next iteration. The foundational start point for such improvements are the sceletonal components of our device - the mechnaical frame from MayTec profiles and thick base plate which build the rigid statical structure and housing for all higher level functionality. Improvement should improve all life-cycle-aspects of the device while maintaining compatibility to the previous device.

Results

First goal was compatiblity - to not compromise defined clincial use cases, auxilary tools and usage procedures, the outline and interfaces were kept the same when possible - in this case the overall heigth and length were kept, while width was increased by 10cm.

Second goal, was the improvement of EMC and water design - sealing and water proofness. Improvements could be achieved by simplification - less complexity, less gaps to worry about.

Overarching theme was simplification - a safe approach by reduction of components and reducing complexity only, no risky innovations. Improved maintainance access to the complex device's many sub-components were the core goal here - while expandability and reduced costs were welcomed secondary benefits.

In summary, for the new v3.2 device iteration an improved maintainability could achieved with reducing the number of frame components by ~20% while keeping the core functionality intact.

Speaker: Jola Klotz (KIT)

10 Towards multiparameter full-waveform inversion for in-vivo ultrasound computed tomography

Full-waveform inversion (FWI) is a powerful reconstruction technique to generate high-resolution models of tissue and bone structure using non-invasive ultrasonic measurements. Sound-speed and reflectivity distributions within the tissue are reconstructed via an iterative data-fitting procedure that models the non-linear relationship between the ultrasonic wavefield and the model parameters. In previous work, we applied FWI to an in-vivo data set acquired with a tri-modal optoacoustic ultrasound imaging platform and reconstruct a high-resolution reflectivity model as well as the speed-of-sound and density distributions within the soft tissue. In this work, we aim to improve the specificity and sensitivity for differentiating between tissue types by additionally inverting for attenuation. Since waveform tomography critically relies on the availability of broadband signals, inverting for attenuation poses a greater challenge than inverting for sound speed or density. Therefore, in a first step towards a multiparameter waveform tomography including sound speed, density, and attenuation, we propose to augment FWI with an amplitude tomography framed as a linearized straight-ray problem. The corresponding linear forward problem expresses the mean attenuation coefficient along the propagation distance as the ratio of the amplitude spectrum of the first-arrival pulse through the previously reconstructed sound speed model and the background water model. The inverted attenuation distribution can then serve as a starting model for a full-waveform amplitude tomography. Because sound speed and attenuation contain complementary information to differentiate malignant masses, merging sound speed, reflection and attenuation images has been shown to visually enhance tumor regions. The fusion of detailed waveform tomography models therefore provides additional benefits in the resolution of speculated masses.

Speaker: Ines Ulrich (ETH Zürich)

11 High-impedance, high-attenuation backing for broadband transducers

To emit short broadband pulses, piezoelectric transducers require a high acoustic impedance, highly attenuating backing material that reduces backside reflections. The current generation of Transducer Array Systems (TAS 3.0) for the KIT 3D USCT III device uses a polyurethane-tungsten composite backing. Although this backing exhibits significant acoustic attenuation, some drawbacks are its relatively low acoustic impedance (Z ~ 5.5 MRayl) and processing difficulties. Compared to polyurethane, epoxy resins are much easier to process, but have lower sound attenuation. However, improved attenuation can be achieved by mixing epoxy with large-grain tungsten powder. This allows epoxy to be used in the backing material, taking advantage of its favorable properties.

Samples of epoxy-tungsten composite backing were prepared by mixing large-grain tungsten powder with epoxy. To avoid precipitation, the epoxy was pre-cured prior to mixing. After final curing, the samples were ground flat to the desired thickness and polished.
This method enabled the production of tungsten-epoxy composites with a tungsten volume concentration of up to 40%, resulting in a high acoustic impedance (Z > 10 MRayl).
The acoustic characterization of the samples was performed in an ultrasonic through-transmission setup. A broadband chirp (0.5 MHz < f < 5 MHz) was used as the excitation signal. The sound velocity of the material was calculated to determine the acoustic impedance of the samples. To calculate the attenuation, the signals were corrected for reflection losses at the media boundaries.

The epoxy-tungsten backing formulation was applied to the next generation TAS 3.1. The frequency and angle-dependent acoustic field of the TAS was measured using a hydrophone. We present results of the characterization of the backing composites as well as initial results of the improved acoustic performance of the new TAS showing an increase in the -10 dB bandwidth of more than 15 %.

Speaker: Patrick Pfistner (Karlsruhe Institute of Technology)

12 Research on Radiated Emission Measurement Method for Medical Ultrasound Diagnostic Equipments

All medical electrical equipment generates electromagnetic radiated disturbances, and the emission levels should comply with the limits required by relevant standards to protect the public radio services in the electromagnetic environment where they are located. Existing international standards only provide general methods and basic principles for measuring radiated emission. For ultrasound diagnostic equipment, including ultrasound tomography equipment, it is not clear under what arrangement or working conditions the maximum emission level can be obtained. Here, we conducted theoretical and experimental studies on various scenarios, including parameter settings, transducer selection, operating modes, and whether to use tissue-mimicking ultrasound phantoms. We found that under the same test conditions, the test results using phantoms were significantly higher than those without phantoms. In addition, different operating modes and transducer configurations can also have a significant impact on the measurement results. Collectively, we propose a specific measurement method for radiated emissions of imaging-type ultrasound diagnostic equipment, which is of great significance for improving the electromagnetic compatibility level of such equipment and ensuring the reproducibility of measurement results.

Speakers: Dr Yang Xu (Y. Xu), Dr Hui Zhang (H. Zhang)

13 Ultrasound Transcranial Simulation Based on Fabry-Perot Resonance

Background
In brain ultrasound imaging, the high impedance barrier presented by the skull is a significant challenge, as it impedes the transmission of acoustic waves into the skull interior. To address this issue, the utilization of a Fabry-Perot resonance-tailoring panel (RTP) placed in front of the skull can be employed to enhance sound wave transmission through the skull by manipulating multiple scattering interactions between the RTP and the skull.
Method
To validate the effectiveness of our method, we conducted two simulations using the acoustic module of COMSOL MULTYPHYSICS 6.1 at a frequency of 500kHz. The simulation domain was filled with water as the background material. The first simulation aimed to demonstrate the transmittance of our method, employing two parallel panels. One panel, with a width of 1.00mm, represented the skull material, while the other, with a width of 0.17mm, represented steel. The gap between the two panels was set at 0.228mm. The second simulation aims to obtain simulation data close to reality, employing a two-dimensional model that mimics the cross-sectional view of the brain, incorporating an annular structure external to the cranial vault to symbolize the presence of RTP.
Result
The results demonstrate that without the implementation of RTP, the ultrasound transmittance is 31.30%. After the implementation of RTP, there is a significant improvement in ultrasound transmittance, reaching 99.95%, and the obtained signal exhibits no distortion.
Conclusion
In brain ultrasound imaging, the transmission of acoustic waves is impeded by the presence of the skull, while this study has substantiated the efficacy of RTP in enhancing the transmission of ultrasound signals. The introduction of RTP enables the overcoming of the high impedance barrier imposed by cranial bones, suggesting the potential application of RTP in future brain ultrasound imaging.

Speaker: Zaituo Li (Huazhong University of Science and Technology)

Tuesday, June 11

Plenary Presentation: Ben Cox

Break

Hardware Development, System Calibration & Phantom Validation

14 Tissue-mimicking phantom for reflection image performance testing of the ultrasound computed tomography instrument

Ultrasound Computed Tomography (USCT) is a radiation free imaging technique that provides acoustic properties information of the human tissue. It commonly provides three image modalities: reflection image, sound speed image and attenuation image. As a newly developed medical instrument, the imaging performance of the USCT should be tested following standard method and provided to the users for product registration and clinical diagnostic reference. Tissue-mimicking phantom is recognized as a standard test tool for quality assurance of medical imaging equipment. We designed and built a universal tissue-mimicking ultrasound phantom for USCT instruments to test their imaging performance in reflection modality. Involved test parameters include spatial resolution (axial and lateral resolution at different radius distance), perimeter and area imaging accuracy, lesion detection capabilities (cyst target, calcific nodules target, tumor target). This phantom contains standard background ultrasound tissue-mimicking hydrogel material with embedded nylon monofilament and lesion-mimicking targets within an enclosed circular truncated cone housing. The housing of the phantom is constructed of upper and lower PMMA plates and plastic acoustic window surrounding at the cuved side face of the truncated cone. The slant angle of the truncated cone side of the housing was designed to simulate real breast shape in water medium and to prevent the primary ultrasound echo transmitted from the circular concave array element. The embedded nylon monofilament and lesion-mimicking targets are designed to be throughout the upper and lower PMMA plates. To keep the structural intensity of the phantom construction, we built three supporting pillar with sound absorption material surroundings between the plates. For the validation of the tissue-mimicking phantom, a 2048-element circular ring array USCT system was used to obtain the tomography image in relection modality. The results demonstrated the imaging performance testing capability of the phantom to USCT system.

Speaker: Di ZHANG (Institute of Acoustics, Chinese Academy of Sciences)

15 A sound speed phantom for hybrid photoacoustic-ultrasound transmission tomography

A composite test object comprising regions with different sounds speeds was developed to evaluate the ultrasound-transmission tomography component of the PAM3 system. The system is capable of hybrid photoacoustic-ultrasound transmission tomography. The test object is made of various materials such as olive oil, copolymer-in-oil and water, and has values of sound speed as expected in breast. The object does not share any inner morphological likeness to the breast other than the outer shape. The different materials used were independently characterized using a reference method to measure speed of sound and acoustic attenuation. We discuss the fabrication, characterization and validation of the test object. We present the images acquired of the test object using the PAM3 system.

Speaker: Esther Bosman (University of Twente)

16 A Calibration Algorithm for a Large Circular Array of Ultrasound Transducers

We consider a planar circular array of a large number of ultrasound emitter-receiver elements. Due to unavoidable inaccuracies of the manufacturing process, the actual positions of the elements slightly differ from the ideal equidistant positions on the circle; also, there exist delays in emitting and receiving the signals. For every emitter-receiver pair available are the accurate enough measurements of the time between the instant of activation of an emitter and the instant of registration of the signal by a receiver, and this is the only information we possess. The inaccuracies mentioned above lead to incorrect evaluation of the actual time-of-flight of the signals and to corrupted resulting ultrasound pictures, so that the numerical values of the misplacements and delays are to be evaluated. This procedure is referred to as calibration.

The existing approaches either recast this problem into a certain convex optimization setup based on use of distances between the elements, or work directly with the coordinates. In the former case, the dimension of the problem (number of variables, size of constraints) becomes huge, which severely complicates data storage and the solution itself. In the latter case, the problem is nonconvex, so that only local minima (i.e., suboptimal solutions) can be found.

In this note we propose a very simple iterative procedure to solve this calibration problem. At every iteration, we first optimize over a part of variables (namely, delays) and then use their updated numerical values to optimize over the rest of the variables (coordinates). With a simple trick, both optimization problems are converted into linear ones leading to moderate-sized linear programs. This makes the overall algorithm really fast even for very large arrays. Preliminary experiments testify to a rather promising performance of the method.

Speaker: Ming Yuchi (Huazhong University of Science and Technology)

17 First results of KIT 3D USCT III system crosstalk analysis

First results of KIT 3D USCT III system crosstalk analysis

Motivation

The KIT 3D USCT III system is built around the idea of SAFT. One ultrasound transducer emits a signal while thousands of others receive - Leading to a system with large number of parallel channels consisting of cables, traces, and connectors. SAFT requires unfocused small transducers necessitating a multi-stage amplification chain, sensitive to crosstalk. As we have to maintain the integrity and fidelity of the signals over the long path until digitization, crosstalk is the natural enemy of this system’s approach.

Method

While electrical crosstalk analysis can be done on a component level, the complete analysis would be a long endeavour and would not represent the final system-level situation. This also neglects non-electrical effects like mechanical propagation. Traditional methods for system-level signal analysis are not applicable, as they require positive SNR. We propose a higher-level crosstalk analysis method, inspired by SAFT, which creates the required SNR by projecting the signals into a 3D spatial domain. We can target emission and receiving paths independently, representing the complete chain from excitation over ultrasound to data.

Results

Our method applied to a typical USCT calibration measurement as emitter detection is shown below. On the left, we can see a good element with a defined emission; On the right, the element is dominated by crosstalk (dead or damaged element) and is visible and can be quantified.

Summary

This in-system method is a fast and complete analysis without the need for extra hardware and manual work steps. It shows potential for hardware debugging by identifying crosstalk paths and components. Digital effects and cabling misconfigurations can also be identified as well. We quantify the worst-case crosstalk of -9dB compared to a typical working signal chain.

Speaker: Michael Zapf (KIT)

MUST2024 v1.08.pptx

18 A serially scalable emission and acquisition system architecture for ultrasound computed tomography

Background:
Most emission and acquisition systems for research and applications in ultrasound computed tomography (USCT) are custom-built and lack scalability in the number of emitting and receiving channels. To address this problem, we proposed a serially scalable emission and acquisition system architecture that is available to build emission and acquisition systems with different channel numbers based on serial control and connections.
Method:
In this architecture, a serially scalable emission and acquisition module (SSEAM) is the core component, which is designed with N channels, consisting of a power board, an emission board, and an acquisition board. Each SSEAM provides a serial clock interface and a serial control interface where clock signals and control signals are serially transmitted and synchronized by the FPGA. By cascading multiple SSEAMs, it is available to construct emission and acquisition systems with a channel number that is an integral multiple of N. Moreover, SSEAM provides a 10-Gigabit Ethernet interface to transfer the acquired data, which can be connected to network cards or high-performance computing cards to complete data post-processing at high speed.
Results:
We constructed emission and acquisition systems with 512, 768, and 1024 channels by cascading 2, 3, and 4 SSEAMs, respectively. Each SSEAM contains 64 emitting channels and 256 receiving channels. All constructed systems can sequentially emit pulses and acquire ultrasound signals. The maximum error of synchronous clock signals is less than 200 ps. The maximum current of each SSEAM is less than 3 A. The maximum rate of 10-Gigabit Ethernet interfaces is 9.8 Gbps.
Conclusion:
The characterization and results indicate that the serially scalable emission and acquisition system architecture based on SSEAMs contributes to constructing an accurate emission and acquisition system with scalability in channel numbers. The constructed emission and acquisition systems are more flexible and can control multiple probes with different element numbers.

Speaker: Quanquan Liu (Department of Biomedical Engineering, School of Life Science and Technology, Huazhong University of Science and Technology)

Lunch

Reconstruction Methods & Modeling: Part 1

19 Coupled Viscoacoustic-Viscoelastic Spectral-Element Modeling for Transcranial Ultrasound

A formulation of the coupled viscoacoustic-viscoelastic wave equation is proposed for modeling the propagation of ultrasound waves in soft tissue-bone systems using the spectral-element method. Including the attenuative effects within the skull is of considerable relevance across a variety of ultrasound applications due to the highly dissipative nature of the trabecular bone. The incorporation of attenuation within both the fluid and solid regions of the domain using the spectral-element method offers a convenient framework for modeling ultrasound propagation in highly heterogeneous media, such as within transcranial ultrasound, given one's ability to explicitly mesh the soft tissue-bone interfaces accurately within the spatial discretization.

A careful choice in where the attenuative terms enter the wave equations allow for the coupled formulation to be solved fully explicitly while simultaneously maintaining the diagonality of the mass matrix, which is one of the key computational properties of the spectral-element method. While the attenuative terms in the elastic medium are incorporated in the stiffness term, attenuation in the acoustic parts of the domain are introduced by modifying the pressure directly. This allows for the system of equations to be solved fully explicitly within the second-order Newmark time-stepping scheme, while avoiding the need to solve a linear system for the interface conditions.

Several numerical examples applied to transcranial ultrasound are presented, particularly within the area of focused ultrasound. Domains with varying degrees of geometric complexity are shown in order to illustrate this technique's efficacy and flexibility.

Speaker: Patrick Marty (ETH Zürich)

20 Interface identification and sound speed reconstruction in layered media: Advancing from synthetic to authentic Ultrasound data

We present a Full Waveform Inversion (FWI) approach designed to simultaneously identify interfaces and reconstruct sound speeds within layered media. This method hinges on synthetic data produced from our forward simulation. By reformulating the inverse problem into a Partial Differential Equation (PDE) constrained optimization, governed by the wave equation, we utilize a gradient descent algorithm to drive optimization. Our results vividly illustrate the efficiency and accuracy of the algorithm in achieving both interface identification and sound speed reconstruction.

In the subsequent section, we will present ongoing progress on performing our inverse algorithm, leveraging experimental measurements provided by our collaborators at the Medical University of Vienna. This may encompass validating the forward model, retrieving the source, and reconstructing the geometrical and physical parameters of such layered media using authentic data.

This is joint work with Peter Elbau, Michael Figl, Otmar Scherzer, and Lukas Zalka.

Speaker: Huidong Yang (Universität Wien)

21 Extending diffraction tomography to focused ultrasound configurations

The inverse acoustic scattering problem refers to the mathematical imaging problem of reconstructing the speed of sound in a medium from a collection of scattered waves. A popular approach is full waveform inversion, which addresses this non-linear inverse problem iteratively without simplifying the underlying mathematical model leading to highly accurate solutions at the expense of increased computational effort.

Diffraction tomography (DT) provides an alternative approach by linearizing the inverse problem under the first Born approximation, enabling efficient computations and hence potential application in medical ultrasound imaging. However, in conventional DT, the incident wave is assumed to be a monochromatic plane wave. This is an unrealistic simplification in medical ultrasound imaging where a transducer typically emits focused beams to a region of interest in the human body.

In this talk, we extend conventional DT by introducing the concept of Gaussian fields of incidence. Herewith, focused beams are modeled, allowing customization by adjusting the beam waist and focal depth. We present a new forward model that incorporates a Gaussian field of incidence and extends the classical Fourier diffraction theorem to the use of this incident field. This focused illumination approach enables new measurement geometries for data generation, based on which we develop reconstruction methods. These are then comprehensively evaluated through numerical experiments.

Speaker: Noemi Naujoks (University of vienna)

Break

Reconstruction Methods & Modeling: Part 2

22 Hybrid Gaussian Beam Pseudo-spectral Method for Efficient Wave Propagation in PAT

Photoacoustic tomography (PAT) is a hybrid imaging technique based on the photoacoustic effect. The PAT forward problem can be modelled as an initial value problem for the free space wave equation. The PAT inverse problem aims to recover an initial pressure from pressure time series recorded at sensors placed outside the region of interest. Despite the advances made in the recent years (parallel interrogation with up to 64 beams), the data acquisition time in state-of-the-art PAT scanners is still a bottle-neck resulting in sparse, limited angle data. The solution of inverse problems with incomplete data necessitate iterative methods involving repeated calls to the forward solver, which is the most compute intensive part of the process. Inspired by the Multiscale Gaussian Beam method proposed by Qian and Ying, we devise an efficient hybrid wave solver, leveraging Gaussian Beams for efficient and highly parallel propagation of high frequency components of the solution, and a pseudo-spectral method for accurate solution of the low frequency components. We discuss the accuracy and performance of our method on an example of solution of the forward problem in PAT.

Speaker: Elliott Macneil (UCL)

23 Uncertainty Estimation in Breast Multiparameter Full Waveform Inversion

Full waveform inversion (FWI) is a reconstruction algorithm recently explored in the field of ultrasound computed tomography (USCT), for high resolution 3-dimensional imaging of the breast. Ultrasound tomography-based imaging, devoid of radiation, can be a safe tool for breast cancer screening and diagnostics. FWI usually inverts for single parameter, typically speed of sound. However, quantitative imaging of other acoustic parameters such as density and attenuation, also can be performed. Such multi-parameter inversions can be especially useful in differentiating various breast tissues, especially to detect any malignancy, leading to more confidence in diagnostics. Nevertheless, the need for extensive computational resources and difficulty in overcoming crosstalk, i.e. separating influence of multiple parameters on the result, decreases its practical applicability.

As a method to improve the single parameter inversion, and to obtain quantitative images of the other acoustic parameters, we propose to make use of linear empirical relationship between speed of sound and density in the inversion process. The linear empirical relationship is used to create intermediate density maps based on the estimated speed of sound map at each iteration and is fed back to inversion process as heterogeneous density map providing prior density information. This heterogeneous density map replaces the constant density value typically employed in the inversion process.

In addition, we estimate the uncertainty associated with these inversions, using an uncertainty quantification (UQ) method based on stochastic variational inference, which was previously proposed for quantifying the confidence of the speed of sound estimate from FWI.

This work uses in-silico and in-vitro phantom datasets to assess the performance of the improved inversion with density feedback and the uncertainty quantification. In addition, the resolution of the 2D-imaging method for given probe centre frequency is quantified using an in-silico resolution phantom with breast tissue acoustic properties.

Speakers: Aparna Harindranath (Imperial College London), Lluis Guasch

Boat tour

A 90min boat tour through the canals that will end at the conference dinner restaurant. You can bring own drinks but no food is allowed on board. The boat will start at
Rederij P. Kooij canal boat tours
Oude Turfmarkt 125, 1012 GC Amsterdam
Make sure to get to this address (Oude Turfmarkt 125, 1012 GC Amsterdam), just entering "Rederij P. Kooij" into google maps will show multiple locations!

Conference Dinner

The conference dinner is held at the restaurant "De Kop Van Oost" (address Zeeburgerpad 1, 1018 AH Amsterdam). All drinks and food are included in the registration.

Wednesday, June 12

Plenary Presentation: Christian Böhm

Break

Reconstruction Methods & Modeling: Part 3

24 Fourier Beamforming for Enhanced Resolution in Ultrasound Computed Tomography

Background:
Ultrasound Computed Tomography (USCT) based on circular arrays provides richer and more detailed information compared to traditional ultrasound imaging, which has attracted extensive attention. The delay-and-sum (DAS) method, along with its related weighted improvement strategies, stands as the mainstream signal beamforming approach in this domain. In the DAS method, the received echo signals are aptly delayed, followed by a summation of these time-aligned signals to enhance the signal at the imaging point. However, the DAS method assumes wave propagation along straight paths, resulting in limited resolution when addressing complex structures in media where multipath propagation exists.
Method:
Addressing these limitations, this study proposes a novel USCT algorithm for circular arrays based on Reverse Time Migration (RTM). This method aims to effectively handling wave propagation in heterogeneous tissue structures. After the Fourier transformation is applied to both the transmitted and received wavefields, by extrapolating the wavefield based on the two-dimensional acoustic wave equation in polar coordinates, our approach obtains the transmitted and received wave fields in the fourier domain at identical spatial locations with a reduced radius, followed by cross-correlation and summation along the frequency axis.
Results and Conclusion:
Experiments on resolution phantom data collected by a circular-array USCT system were conducted. The lateral and axial Full Width at Half Maximum (FWHM) values of the chosen points are calculated to verify the performance. Compared with the classical Coherence Factor-weighted DAS (CF-DAS) method, our approach can offer a 15% higher FWHM value than CF-DAS method. Therefore, our approach has the advantage of improving image resolution in USCT through more accurate wavefield reconstruction. Additionally, it is noteworthy that the results demonstrate that our method can effectively suppress the artifacts associated with multiple reflections. Moreover, leveraging parallel processing on GPUs, this algorithm shows promising potential for clinical applications.

Speaker: Xiang Zhou (HUAZHONG UNIVERSITY OF SCIENCE AND TECHNOLOGY, COLLEGE OF LIFE SCIENCE & TECHNOLOGY)

25 Determine power law parameters for ultrasound attenuation imaging

Ultrasound attenuation maps are an important imaging modality of medical ultrasound tomography. Many approaches however only put focus on the attenuation at a specific or dominant frequency with small bandwidth or the broad band attenuation of a signal with large bandwidth. Yet, attenuation by tissue is typically considered to be frequency dependent. In the literature this is modelled linearly or with a power law (Attenuation = $ a \cdot f^y$ in $\mathrm{\frac{dB}{cm}}$, where $a$ is attenuation coefficient in $\mathrm{\frac{dB}{MHz^y \cdot cm}}$, $f$ is frequency in $\mathrm{MHz}$ and $\mathrm{y}$ is attenuation exponent).

Using the approach of the power law attenuation, we developed a method to determine the according parameters (a and y) from measured data of KIT’s 3D USCT III with broadband signals (0.5-5 MHz). The individual broadband attenuation signals are being windowed and transformed into the Fourier domain to calculate an attenuation value for sub-bands of the available bandwidth. Subsequently an attenuation map is reconstructed for each sub-band. These attenuation maps are used to perform a parameter fit for each voxel to determine the two parameters a and y.

The method has been applied successfully on simulated 3D data using a ray based simulation suite to validate the approach (RMSE of a = 0.294 $\mathrm{\frac{dB}{MHz^y \cdot cm}}$, RMSE of y = 1.361). Furthermore the k-wave toolbox has been used to test the concept with 2D simulated data (RMSE of a = 0.0316 $\mathrm{\frac{dB}{MHz^y \cdot cm}}$, RMSE of y = 0.2197). In addition results with experimental data will be presented.

Speaker: Clemens Feucht (Karlsruhe Institute of Technology)

26 Full 3D ultrasound tomography: current status at KIT

Ultrasound tomography (USCT) is an exciting new technology with several active research groups investigating new algorithms, devices and applications worldwide. To fully utilise the 3D interaction of the ultrasound fields with the object to be imaged, we are focusing our research on 3D USCT systems. We have realised a pseudo-randomly sampled hemispherical 3D USCT device (3D USCT III) with 2304 transducers with nearly spherical wave fronts for transmission and reception. Due to the large amount of acquired data and the large 3D space that needs to be reconstructed, image reconstruction is performed using straight and bent ray algorithms, while the more advanced paraxial and full waveform inversion reconstructions are currently under development.

In this summary, the design of the system and initial results with phantoms, volunteers and the ongoing clinical trial are presented. Wire, sphere and gelatin-based phantoms were used to access resolution and contrast of all modalities, and accuracy of speed of sound and attenuation reconstructions. Volunteer images showed plausibility of reconstructions in comparison to MRI and greatly increased field of view, e.g. visibility up to chest muscle in reflection images.

Speaker: Nicole Ruiter (Karlsruhe Institute of Technology)

27 Numerical methods for weakly nonlinear ray tracing approximations in HIFU

Focused ultrasound is used in a therapeutic treatment (HIFU) and uses ultrasound waves to non-invasively destroy malignant cells inside the human body. The technique works by sending a high-energy beam of ultrasound into the tissue using a focused transducer. Numerically modelling HIFU presents a problem due to nonlinear effects leading to the formation of harmonics of the source frequency. Each harmonic requires a finer grid to resolve, rapidly increasing computational complexity.
In prior work we unified two ray tracing methods using weakly nonlinear ray theory. The first has the ray equations identical to those from linear ray theory while the amplitude equation is a nonlinear transformation of the Burgers’ equation. In the second method, the Eikonal and transport equations are coupled which results in ray trajectories which depend on the amplitude. This presentation explores numerical comparisons of the two methods with the pseudo-spectral solution of the Westervelt equation.

Speaker: Matt Foster (University College London)

28 Computed Ultrasound Tomography in Echo Mode – current status

Computed Ultrasound Tomography in Echo Mode (CUTE) maps the tissue’s speed-of-sound (SoS) using conventional handheld limited-aperture echo ultrasound (US) probes. The technique consists of a sequence of steps that can be performed in real time: beamforming of US images under varying transmit- and receive steering angles, detection of echo shift between different angle combinations, SoS inversion from the echo shift data. Recently we have adapted CUTE to also image the US attenuation coefficient.
After a recap of the basic principles and the recent clinical results, this talk will focus on discussing critical aspects of CUTE: (i) the role of regularization for quantitative imaging. Despite the ill-posedness of the SoS inversion, quantitative imaging is in principle feasible subject to an appropriate prior. (ii) the challenge of imaging real tissue. Wave aberrations at short scale heterogeneities of SoS degrade the echo shift data, leading to artifacts and – more importantly – reduced quantitative accuracy. (iii) artificial neural network based SoS inversion. This may be a clever alternative to a full-wave inversion to account for aberration effects, however, care needs to be taken to appropriately design the training data.

Speaker: Michael Jaeger (Institute of Applied Physics, University of Bern)

Lunch

Plenary Discusion

Open discussion on questions that are of wider interest to the UST community, such as
- The role of the MUST workshop in relation to other (bigger) conferences like SPIE Medical Imaging: Ultrasonic Imaging and Tomography: What MUST do we need and where should it be next?
- What kind of open data sets / software would be useful to have?
- What are the big challenges of UST we should work on? From a methodical / engineering / clinical / commercial perspective
You can already think about these points in advance and please bring any questions you would like to discuss with everyone or any announcements you'd like to make!

Break

Machine Learning

29 Breast Reflection Ultrasound Computed Tomography Reconstruction with Self-Supervised Blind Sequence Network

Reflection ultrasound computed tomography (RUCT) is emerging as an essential tool for clinical breast cancer screening. However, a persistent challenge in ultrasonic imaging lies in the degradation of image quality due to local sound speed variations in breast tissue and random noise in the circuitry. RUCT imaging is based on the classical delay and sum (DAS) algorithm. Its pixel value is directly determined by the Time of Flight (TOF), i.e., the RF data delay time from the transmitting element to the target and then to the receiving element under unrealistic uniform sound speed model. On the other hand, only a bandpass filter is manually designed to alleviate electrical noise. In this paper, we propose a deep learning method, termed the Self-Supervised Breast RUCT Reconstruction (SSBRR) framework, tailored specifically for improving RUCT through RF data processing. Our framework focuses on the challenge of accurately locate RF data delay from its subsequences and explore potential benefits of mitigating destruction through the slackness. Then, the estimation of RF data delay is significantly contributed by capturing spatial latent consistency across the receiving arrays. To evaluate the effectiveness of our proposed method, we employ well-established metrics, including Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Metric (SSIM). Our experiments demonstrate promising results, with our method achieving an average PSNR of 20.892 and an average SSIM of 0.792, particularly under conditions of extreme sparsity in transmission. The experimental analyses conducted attest to the superior performance of our framework compared to competing enhancement strategies.

Speaker: Lei He

30 Coordinate-based implicit neural representation for full waveform inversion in ultrasound computed tomography

Background:
Sound speed imaging is an important characteristic of ultrasound computed tomography (USCT). Full waveform inversion (FWI) is regarded as the most promising algorithm for high-resolution sound speed imaging. However, FWI may encounter the phenomenon of cycle-skipping, which means falling into local optimal solutions. Implicit neural representation (INR) is a popular technique that has been developed recently. The technique has demonstrated impressive results in tasks such as super-resolution and image generation. This study explores the application of INR to the field of sound speed imaging.
Method:
We present a multilayer perceptron (MLP) network that receives the coordinates of image pixels as inputs and predicts the sound speed values at the corresponding positions. This network is employed to create a sound speed image. The generated sound speed image is subsequently used in the forward process to simulate signals, followed by the backward process in which the sound speed is iteratively refined through the adjustment of the MLP network parameters.
Results:
We conduct a simulation experiment using a numerical model of a breast to verify the effectiveness of the algorithm. The simulation experiments are performed with a circular array of 64 transmitters and 256 receivers at a toneburst wave with 350 kHz center frequency. The baseline method is based on the least-square FWI. The root mean square error of the results of the proposed method was 5.9 and 1993.3 for the baseline.
Conclusion:
The results demonstrate that the INR network is able to act as an implicit regularizer for the FWI algorithm, thus preventing it from falling into a locally optimal solution. Therefore, it is believed that the proposed method offers a feasible alternative to the current sound speed imaging in USCT.

Speaker: Weicheng Yan (Huazhong University of Science and Technology，College of Life Science and Technology)

31 Neural Network-Based Data Recovery Using Spiral Sub-Sampling for Ultrasound Computed Tomography

Ultrasound computed tomography (USCT) has the potential for clinical applications due to its standardized operations and multi-modality. However, obtaining high-quality images requires a complete dataset including all transmitting-receiving pairs, resulting in a time-consuming scanning process and substantial data-processing demands. The limitation restricts the clinical applications of USCT. Reconstructing images directly from sub-sampled radio-frequency (RF) data leads to diminished image quality. To address this issue, we propose an efficient spiral sub-sampling strategy for sparse data acquisition and a data recovery approach based on convolutional neural network (CNN).

The spiral sub-sampling strategy selects receiving channels symmetrically, centering around the transmitting array. Specifically, for the n-th transmitting event, echo data are collected from the n, n+m, n+2m, and subsequent elements. The uniform sub-sampling strategy employs fixed channels for reception. Specifically, for every transmitting event, echo data are collected from the m, 2m, 3m, and subsequent elements. To validate the approach, six sets of complete RF data from the human leg were acquired, the first five sets were used to train the CNN, and the remaining was used for testing. We conducted experiments using the CNN to recover data sub-sampled by a factor of 4× or 8×. To quantitatively show the advantages of the proposed method, we used structure similarity (SSIM) and peak signal-to-noise ratio (PSNR) as metrics to assess the quality of reconstructed images.

In 4× Rx sub-samping experiments, the uniform and spiral methods achieved SSIM and PSNR values of 0.07, 0.08, and 7.16 dB, 6.52 dB higher than the input, respectively. In 8× Rx sub-samping experiments, the uniform and spiral methods achieved SSIM and PSNR values of 0.02, 0.07, and 0.46 dB, 5.88 dB higher than the input, respectively. The results show that for higher downsampling rates, our proposed method demonstrates the potential to utilize the data redundancy in the transmit-receive plane.

Speaker: Hantao Yang (Department of Biomedical Engineering, Huazhong University of Science and Technology)

32 ASPIRE: Iterative Normalizing Flows for Fast Transcranial Ultrasound with Uncertainty Quantification

Transcranial ultrasound computed tomography using Full-Waveform Inversion presents a unique challenge due to the non-linear physics and the computational expense of wave physics. We address this challenge with a probabilistic framework that learns to sample the Bayesian posterior of brain parameters that match the data. To scale to realistic parameter sizes, we use normalizing flows and preprocess the raw waveforms with a physics-based summary statistic. In order to mitigate non-convexity, we propose a novel approach that alternates between normalizing flow training and improving the summary statistic using the current estimate from the normalizing flow. We denote our proposed method, ASPIRE: Amortized posteriors with Summaries that are Physics-based and Iteratively REfined. Our evaluation shows that our method can accurately image through the skull while maintaining low online costs at four PDE solves, compared to the hundreds of PDE solves used in traditional approaches. To demonstrate the accuracy of our image reconstruction and its uncertainty quantification, we compare our method against a technique from the literature and find that our approach provides more accurate reconstructions, is faster, and offers better-calibrated uncertainty.

Speaker: Rafael Orozco (Georgia Institute of Technology)

Break

Plenary Presentation: Richard Lopata