Who We Are

Ultrasound imaging is a widely used diagnostic tool for liver diseases, yet conventional B-mode imaging has limitations in detecting diffuse liver pathologies at early stages. This dissertation explores the application of H-Scan analysis on handheld ultrasound (HH-US) radiofrequency (RF) data to assess diffuse liver diseases, particularly liver steatosis. The study evaluates the feasibility of H-Scan scatter ratio as a quantitative imaging biomarker, comparing results with controlled attenuation parameter (CAP) values obtained from transient elastography (FibroScan®).

Using a large patient cohort, RF data were recorded with high-frequency linear and low-frequency abdominal HH-US probes. The study demonstrates that the H-Scan scatter ratio significantly differentiates between patients with severe steatosis (S3) and those without steatosis (S0). Furthermore, a strong correlation between scatter ratio and CAP values was observed. These findings suggest that quantitative H-Scan analysis, integrated with portable ultrasound devices, could enhance early-stage liver disease diagnostics and offer a cost-effective alternative to conventional elastography techniques.

Abstract Chronic liver diseases affect ~1.5 billion people worldwide and cause 3.5% of all deaths. Hepatic steatosis can progress to irreversible damage, making early detection crucial. Non-invasive assessment has been challenging, but quantitative ultrasound-based methods are gaining importance. This study aimed to develop a quantitative ultrasound parameter to assess liver steatosis using transient elastography’s controlled attenuation parameter (CAP) as a reference. Raw ultrasound radiofrequency (RF) data from handheld devices were analyzed. The main hypothesis proposed a strong correlation between ultrasound attenuation recorded with a handheld linear transducer and CAP values. Between November 2022 and March 2023, raw ultrasound data were collected from patients undergoing ultrasound and Fibroscan examinations using Clarius C3HD3 and L15HD3 transducers. Three-second recordings were processed using a Python script. A non-frequency-specific attenuation parameter (AP) and a frequency-specific AP were calculated via short-time Fast Fourier Transformation (stFFT). Pearson and Spearman correlation coefficients were determined, and Fishers-z tests compared correlations. Data from 135 participants were analyzed. Pearson correlation coefficients between non-frequency-specific AP and CAP were 0.634 (C3) and 0.694 (L15). For frequency-specific AP, correlations were 0.693 (C3) and 0.712 (L15). The correlation between frequency-specific AP and CAP was not significantly higher than that of non-frequency-specific AP. Strong correlations between ultrasound APs and CAP values suggest handheld ultrasound devices can quantify liver steatosis. However, frequency-specific APs require further investigation. Validation with MRI-PDFF is recommended.

This thesis focuses on the comparative analysis of the transmission behavior of piezoelectric and capacitive micromachined ultrasonic transducers (CMUTs) for ultrasound imaging, with an emphasis on nonlinear behavior. The objective is to evaluate and compare the nonlinear transmission properties of these transducers using finite element (FE) simulations and experimental measurements. The Swept-Sine Technique (SST) is applied to quantify nonlinearities in the transmission behavior at different operating points of the transducers.

The results reveal differences in the nonlinear transmission behavior between piezoelectric and CMUT transducers, with CMUTs exhibiting more pronounced nonlinear behavior under similar conditions. The study also examines the effects of varying bias and peak-to-peak voltages on the nonlinear characteristics of the transducers. The findings from the FE simulations largely align with the experimental results, providing insights into the factors influencing transducer nonlinearities. The thesis concludes that CMUTs exhibit stronger nonlinear behavior than piezoelectric transducers, which may impact their application in ultrasound imaging. Additionally, SST has proven to be an effective method for quantifying nonlinearities in the transmission behavior of transducers.

Ultrasonic transducers primarily utilize piezoelectric composite materials, which consist of an active piezoelectric phase and a passive polymeric phase. Single-crystal PMN-PT is a promising material for transducers due to its excellent piezoelectric properties. However, its brittleness presents challenges in micromachining, typically performed using a dicing saw. This thesis investigates the structuring of PMN-PT using ultrashort laser pulses for the fabrication of 1-3 piezocomposites and compares it to conventional dicing saw processing. Theoretical models assess the influence of the processing-induced wall angle on transducer properties, revealing a significant decrease in the coupling factor for angles below 87°. Experimental studies identify optimal laser parameters for structuring PMN-PT, and samples are processed using both laser pulses and a dicing saw for comparison. The results demonstrate that 1-3 piezocomposites for 5 MHz can be successfully fabricated using ultrashort laser pulses, though limitations include processing time and a wall angle.

Sonography is one of the most important non-ionizing, real-time imaging modalities used in medical applications today. The current limitations in image quality and penetration depth could be significantly improved by using long or continuous coded waveforms, which have the potential to enhance the signal-to-noise ratio (SNR). However, these waveforms are associated with considerable loss-induced heating of the transducer, which has not been fully explored in the literature. This thesis investigates the thermal behavior of piezoelectric transducers under continuous excitation across a wide frequency range of 1–20 MHz. Thermal and vibration measurements are performed to understand the system and dominant loss mechanisms. Additionally, a finite element (FE) model is developed to predict the surface temperature rise of the transducer and compared with experimental results. The findings indicate that both thermal and vibrational behavior are highly frequency-dependent, with maximum heating occurring near resonance. Mechanical losses were identified as a major source of heating in resonance conditions. The simulation model showed strong deviations at high temperature increases, likely due to the limited thermal boundary condition adjustments in the FE software. Overall, this study contributes to a deeper understanding of the effects of continuous broadband excitation on piezoelectric transducers. Future work should include a larger number of transducers with lower manufacturing tolerances and variations in backing and matching layers to confirm and expand upon the observed relationships.

Two-photon lithography offers significant potential for the high-resolution structuring of matching layers, which could greatly enhance ultrasound diagnostics in the future. The design of such layers requires knowledge of the acoustic impedance of the polymers used. However, materials suitable for two-photon lithography are typically available only in small volumes, and their acoustic impedances are not well-documented, as they depend on manufacturing parameters. The goal of this thesis is to develop and apply a measurement method suitable for characterizing samples with minimum edge lengths of 200 µm, since conventional measurement methods are not designed for such small samples. This study investigates a reference measurement approach as a suitable method. Two measurements are conducted under identical conditions: one on a sample with known impedance and another on the test sample. By comparing both measurements, the unknown impedance can be determined. Various factors affecting accuracy, such as environmental temperature and surface properties of the samples, are analyzed and, if possible, compensated for. The resulting measurement accuracy exceeds 95%, with a standard deviation below 0.05 MRayl. The developed method allows for both an overall impedance determination and pixel-wise evaluation, enabling topographical representations. The approach is applicable to both microstructures and macroscopic samples with minimal surface condition restrictions. Despite surface roughness, the measurement remains valid due to the application of a threshold. However, factors such as measurement frequency and applied amplification influence the process, and further research is suggested to enhance its general applicability.

This thesis explores the design and evaluation of a handheld ultrasound transducer, focusing on ergonomic and usability improvements for medical professionals. Despite being the most commonly used imaging technology in medicine, ultrasound devices have seen limited innovation, particularly in ergonomic design. The research, conducted within the Hybrid Echo research group at the EKFZ for Digital Health, aims to develop a new ultrasound probe that combines traditional piezoceramic transmitters with novel CMUT (Capacitive Micromachined Ultrasonic Transducer) receivers to enhance image quality. The study evaluates ergonomic parameters such as size, weight, and center of mass, and develops new probe designs through iterative prototyping and user testing with medical professionals. Results indicate that smaller, lighter probes with a patient-side center of mass are preferred, and that the new designs are generally accepted, with some users even preferring them over conventional probes. The modular design of the probe allows for adaptability to different use cases, offering potential for future technological integration. The findings suggest that ergonomic improvements can significantly enhance the usability and acceptance of ultrasound devices in clinical settings.

It is assumed that the hybrid use of piezoelectric elements and capacitive micromachined ultrasonic
transducer (CMUTs) elements result in an increased signal SNR and bandwidth in ultrasound
imaging. However, classical imaging techniques can not be implemented with such
hybrid configurations. Thus inverse problem strategies for multistatic acquisition configurations
are investigated for image formation in these settings. However, such methods go along
with an increase in computational overhead that has to be accomplished.
This work aims to compare the image reconstruction speed for Delay-and-Sum – a time domain
imaging method , which is well known in the ultrasound community and a k-space based
algorithm published in the context of radar imaging by Alvarez et al. [´Alv+14]. A model for
ultrasound wave propagation is described and implemented as a simulation. Both algorithms
are implemented in C++. The Implementations are validated using synthetic data produced by
the simulation. A rigorous derivation of the k-space algorithm is given and time complexities
are determined regarding the Random Access Machine (RAM) computation model. The time
complexities were validated by runtime measurements both algorithms. Several advantageous
properties regarding the execution time of the k-space method compared to Delay-and-Sum
are observed. However, it is also pointed out, that further research has to be conducted in
order to apply the k-space method to a hybrid imaging setup.

Ultrasound imaging is a key diagnostic tool in clinical practice, relying on the reflection of sound waves at tissue boundaries. Conventional ultrasound imaging techniques, while widely used, are limited in terms of image resolution and frame rate. To overcome these limitations, Multiple Input Multiple Output (MIMO) Synthetic Aperture (SA) imaging has emerged as a promising alternative. This technique utilizes multiple independent transmitting and receiving elements simultaneously, improving imaging performance.

This study investigates suitable waveforms for MIMO-SA ultrasound imaging, focusing on optimal autocorrelation and cross-correlation properties. A literature review identified two key multiplexing methods: Frequency Division Multiplexing (FDM) and Code Division Multiplexing (CDM). Various spreading codes and modulation schemes were analyzed, including pseudorandom sequences and Constant Amplitude Zero Autocorrelation (CAZAC) sequences. A comparative evaluation was performed using different signal metrics and a two-dimensional point scatterer model.

The results demonstrate that OFDM-modulated Zadoff-Chu sequences, a subset of CAZAC signals, exhibit the best characteristics for MIMO-SA imaging. These findings contribute to the advancement of high-resolution, high-frame-rate ultrasound imaging, with potential applications in real-time medical diagnostics.

UNDER CONSTRUCTION

The utilisation of non-invasive Ultrasound (US) methods in an environment capable of real-time image reconstruction is of high interest for various medical fields. Widely adopted US handheld devices can further benefit from the low costs and the portability. With the emergence of more computationally intensive US imaging methods, the real-time aspect can be lost, resulting in the usage of external hardware. In this work, a Fourier-based imaging (FBI) algorithm will be implemented and optimized on a Snapdragon 8 (S8), to analyse the extent to which the System-On-Chip (SoC) is able to provide real-time execution. To profit from the inherent parallelism of the
integrated Graphics Processing Unit (iGPU), multiple optimizations techniques were applied to reduce the runtime of the algorithm. While evaluating the performed optimizations on several sets of simulation data, it could be shown that the defined real-time constraint is only met for certain parts of the FBI algorithm.

The efficient computation of the Fast Fourier Transform (FFT) is crucial for real-time medical ultrasound imaging, particularly in Fourier-based imaging (FBI). This thesis investigates the acceleration of FFT on the Xilinx Versal VCK190 System-on-Chip (SoC), leveraging its AI Engine architecture for parallel processing. The Cooley-Tukey algorithm is employed to optimize FFT execution across multiple AI Engine tiles, with a focus on reducing memory bottlenecks and maximizing computational throughput.

The study demonstrates that while FFT computation itself meets real-time constraints, data streaming between AI Engines remains a critical performance bottleneck. To address this, an optimized dataflow architecture is proposed, integrating memory-efficient buffering and packet-based data transmission. The results highlight the potential of AI-accelerated SoCs for high-speed signal processing, with implications beyond medical imaging, including radar and mobile communications. This work provides key insights into optimizing large-scale FFT computations in resource-constrained environments.