In progress – view active session
Conference 13412
Ultrasonic Imaging and Tomography
18 - 20 February 2025 | Palm 2
View Session 
Tuesday Morning Keynotes
18 February 2025 • 8:30 AM - 10:00 AM PST | Town & Country B/C
Session Chairs:
Jhimli Mitra, GE Research (United States), Christian Boehm, ETH Zurich (Switzerland)
View Full Details: spie.org/tuesday-morning-keynotes
8:30 AM - 8:35 AM:
Welcome and introduction
8:35 AM - 8:40 AM:
Award announcements
- Robert F. Wagner Award finalists for conferences 13406 and 13412
- Image Processing Student Paper Award
13406-504
Integrating computational approaches to unravel alzheimer's disease and co-pathologies: a biomarker-driven approach to precision medicine
(Keynote Presentation)
18 February 2025 • 8:40 AM - 9:20 AM PST | Town & Country B/C
Show Abstract +
Early detection and intervention in neurodegenerative diseases hold the potential to significantly impact patient outcomes. This presentation will explore the development of multi-disciplinary and multi-modality biomarkers to identify individuals at risk and monitor disease progression. By combining advanced imaging techniques, such as MRI, and PET, with fluid biomarkers, we aim to detect subtle changes in brain structure and function that precede clinical symptoms. These biomarkers could serve as powerful tools for early diagnosis, enabling timely intervention and potentially delaying disease onset. Furthermore, by identifying individuals at highest risk, we can optimize the design of clinical trials and accelerate the development of effective therapies. Ultimately, our goal is to improve the lives of individuals with neurodegenerative diseases through early detection, precise diagnosis, and targeted treatment.
13412-505
Wearable ultrasound technology
(Keynote Presentation)
18 February 2025 • 9:20 AM - 10:00 AM PST | Town & Country B/C
Show Abstract +
The use of wearable electronic devices that can acquire vital signs from the human body noninvasively and continuously is a significant trend for healthcare. The combination of materials design and advanced microfabrication techniques enables the integration of various components and devices onto a wearable platform, resulting in functional systems with minimal limitations on the human body. Physiological signals from deep tissues are particularly valuable as they have a stronger and faster correlation with the internal events within the body compared to signals obtained from the surface of the skin. In this presentation, I will demonstrate a soft ultrasonic technology that can noninvasively and continuously acquire dynamic information about deep tissues and central organs. I will also showcase examples of this technology's use in recording blood pressure and flow waveforms in central vessels, monitoring cardiac chamber activities, and measuring core body temperatures. The soft ultrasonic technology presented represents a platform with vast potential for applications in consumer electronics, defense medicine, and clinical practices.
Break
Coffee Break 10:00 AM - 10:30 AM
Session 1:
Ultrasound Image Processing and Analysis
18 February 2025 • 10:30 AM - 12:30 PM PST | Palm 2
13412-1
18 February 2025 • 10:30 AM - 10:50 AM PST | Palm 2
Show Abstract +
This study validates super-resolution ultrasound imaging using erythrocytes (SURE) against state-of-the-art micro-CT data of rat kidneys. By comparing SURE images and micro-CT scans at a 5 μm voxel size, the research confirms SURE's ability to depict blood vessels with diameters below 50 μm. The study employs the SURE-Hankel method for detailed analysis, revealing consistent visualization of cortical radial vessels. Discrepancies between the images suggest the need for further investigation into how vessel depiction quality varies with position in the ultrasound focus plane.
13412-2
18 February 2025 • 10:50 AM - 11:10 AM PST | Palm 2
Show Abstract +
Transthoracic echocardiography (TTE) is a widely used diagnostic tool for cardiac diseases, requiring accurate
classification of its standard views for downstream tasks using TTE data. Traditional methods often rely on random or fixed video frame sampling, leading to misaligned training data. This study introduces a novel deep learning-based approach that leverages pixel-embedded ECG signals for multi-phase image sampling, targeting key cardiac phases—systole, diastole, and R-wave peak. Using a cardiac cycle detection algorithm, we extract physiologically relevant frames from TTE videos and train separate models, reducing computational demand while improving classification accuracy. Our final model employs an ensemble of EfficientNet-B2 classifiers trained on three phase-specific frames of 6,512 cardiac cycles from 11 TTE views, achieving a mean accuracy of 92.4% and an AUC of 99%. Results demonstrate that view classification accuracy depends on the cardiac phase, with optimal predictions varying based on the structural differences observed at different points in the cycle. This work underscores the potential of ECG-assisted classification to streamline automated TTE analysis.
13412-3
18 February 2025 • 11:10 AM - 11:30 AM PST | Palm 2
Show Abstract +
Early breast cancer detection significantly influences patient outcomes. Dynamic Contrast-Enhanced Ultrasound
(DCE-US) has shown promise in early detection by visualizing tumor vascularity and perfusion dynamics in real-time. This study evaluates the efficacy of DCE-US in a transgenic mouse model that mimics human breast
cancer progression using a VEGFR2-targeted microbubble contrast agent. By omitting traditional ultrasound
burst pulses to remove unbound tracer-laded microbubbles and analyzing pre-pulse data with Non-Negative
Matrix Factorization (NMF), we successfully differentiated between tumor-specific and non-specific binding,
thus enhancing cancer tissue identification. Our findings support the potential of NMF in DCE-US without any
need for a pulse, with significant implications for clinical application.
13412-4
18 February 2025 • 11:30 AM - 11:50 AM PST | Palm 2
Show Abstract +
Ultrasound imaging is widely used for its cost-effectiveness, non-invasiveness, and real-time capabilities in diagnostics and image-guided procedures. However, the direct skin contact involved in ultrasound image acquisition induces deformations in the tissues.
In certain scenarios, such as elastography or freehand 3D ultrasound imaging, the relative deformations between successive ultrasound images must therefore be estimated or corrected. This is a non-rigid image registration operation. Existing methods are often validated on limited test datasets, hindering reproducibility and comparisons. To address this gap, we introduce a publicly available dataset with controlled deformations on a phantom, including B-mode images, RF signals, probe indentations, contact force, and phantom material mechanical parameters. This dataset serves as a valuable resource for validating various registration and elastography methods, offering a standardized platform for performance evaluation and comparison among research groups. The article outlines the experimental protocol, the mechanical system for probe maintenance and indentation, force measurement, and the phantom creation process.
13412-5
18 February 2025 • 11:50 AM - 12:10 PM PST | Palm 2
Show Abstract +
This study presents a novel sensor-free force estimation technique for automatic control of contact force in robotic ultrasound scanning. By employing a deep learning model that combines Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) architectures, the system estimates force differences using sequences of ultrasound images, bypassing the need for complex sensor installations. Experiments conducted without force sensors demonstrated the model's accuracy, achieving a mean squared error (MSE) of less than 0.6 N². This approach simplifies the robotic ultrasound setup and enhances imaging performance.
13412-6
18 February 2025 • 12:10 PM - 12:30 PM PST | Palm 2
Show Abstract +
The prevalence of medical imaging in diagnosing structural heart diseases (SHDs) has increased the use of indices like global longitudinal strain (GLS) to assess cardiac dysfunction. However, GLS, often estimated using B-mode echocardiography, does not capture the complexity of cardiac motion and is challenged by the rapid heart motion in small animal models like mice. We hypothesize that four-dimensional (4D) ultrasound imaging with a high-frequency transducer can provide a complete spatiotemporal description of cardiac motion. In this study, following the reconstruction of a mouse-specific left ventricle (LV), the iterative closest point algorithm, we estimated 4D displacements and strains. The proposed methodology was validated against speckle-tracking-derived displacements in synthetic images. Our methodology offers a rigorous protocol to quantify spatiotemporal cardiac motion, promoting the adoption of regional strain markers in clinical settings.
Break
Lunch Break 12:30 PM - 1:40 PM
Session 2:
Image-Guided Procedures, Robotic Interventions, and Ultrasonic Imaging/Tomography: Joint Session with Conferences 13408 and 13412
18 February 2025 • 1:40 PM - 3:00 PM PST | Town & Country D
Session Chair:
Jessica R. Rodgers, Univ. of Manitoba (Canada)
13412-7
18 February 2025 • 1:40 PM - 2:00 PM PST | Town & Country D
Show Abstract +
In this presentation, we present a modular ultrasound (MODUS) framework, allowing clinicians to have an ultrasound array in arbitrary aperture size and on flexible surface curvatures, which can maintain effectiveness regardless of patient age and physical condition.
13408-21
18 February 2025 • 2:00 PM - 2:20 PM PST | Town & Country D
Show Abstract +
Real-time intraoperative localization during breast cancer surgery is essential to ensure complete tumor resection. Unfortunately, limitations in existing single-modality devices result in high rates of revision surgeries. This research details the design and fabrication of a novel hybrid imaging system that mechanically couples a unique type of focused gamma probe with an ultrasound transducer to simultaneously acquire anatomical and functional images in real time. Compared to existing literature, phantom studies of our radio-ultrasound guided system revealed a substantial improvement in resolution (almost an order of magnitude better) while maintaining high sensitivity. The complementary information provided by precisely visualizing and quantifying ‘hot’ radiolabeled tissues within the anatomy is expected to improve the accuracy of breast cancer surgery, leading to improved patient outcomes.
13412-8
18 February 2025 • 2:20 PM - 2:40 PM PST | Town & Country D
Show Abstract +
Midline shift (MLS) is a diagnostic marker for brain pathologies such as intracranial hemorrhage and traumatic brain injury, requiring emergent treatment. CT and MRI are the gold standards for MLS measurement but prevent repeated, short-interval monitoring and are associated with significant risks when transporting critically ill patients. This feasibility study investigates a freehand 3D ultrasound (US) system for bedside MLS measurement. Using electromagnetic position trackers on both the US transducer and the subject, we reconstructed 2D B-mode ultrasound images into a 3D volume. We scanned a healthy subject through bilateral transtemporal windows and measured MLS on the reconstructed volume. Our system achieved a measurement error comparable to existing sonography-based methods, fitting clinical requirements, and proving its feasibility for clinical use. Our approach reduces operator-skill dependency and provides a rapid, non-ionizing solution for continuous, short-interval monitoring of MLS, potentially enhancing clinical decision-making in critical care and emergency field diagnosis.
13408-22
18 February 2025 • 2:40 PM - 3:00 PM PST | Town & Country D
Show Abstract +
Smaller, affordable ultrasound equipment is widely used, but image quality issues require specialized training. Augmented reality (AR) enhances training but faces challenges in development tools, real-time tracking, and costs. The proposed system utilizes Meta Quest3, interacting with the Clarius L15, an ultrasound simulator. The AR platform employs Touch Plus controllers to track the US probe and biopsy needle, allowing the physician to maneuver the probe freely regardless of movement. To evaluate this method's viability, the accuracy and precision of the Touch Plus controllers in instrument tracking were studied and compared to a commercial electromagnetic tracking system. Experiments showed that the proposed strategy achieved an accuracy of 0.68°±0.55° and a precision of 0.36°±0.03° in 12 different orientations and positions. The results demonstrated the proposed approach's effectiveness in tracking instruments without external markers in various orientations and positions, indicating its potential for clinical practice.
Break
Coffee Break 3:00 PM - 3:30 PM
Session 3:
Imaging Algorithms and Reconstruction Techniques
18 February 2025 • 3:30 PM - 5:30 PM PST | Palm 2
13412-9
18 February 2025 • 3:30 PM - 3:50 PM PST | Palm 2
Show Abstract +
Multiaxial transducer technology uses orthogonal electric fields in the same transducer to detect with high precision the angle of incidence of a front wave using only a single detector. In this communication, we present the foundation for methods for ultrasound computed tomography (UCT) based only on direction of arrival (DOA) information and bent-ray theory, and a numerical demonstration of UCT reconstruction using this approach. Experimental DOA was calculated with a numerical solution of the viscoelastic equation and a method for directivity estimation using the phase map of pressure fields. A full wavefront ray tracer was used as the forward model during the inversion procedure. Results indicated that reconstruction of two phantoms, using fundamental ultrasound frequencies of 300 kHz and 600 kHz, was feasible using only DOA information and provides advantages over conventional methods. To the best of our knowledge, this is the first numerical demonstration that UCT image reconstruction is entirely possible using exclusively directivity information.
13412-10
18 February 2025 • 3:50 PM - 4:10 PM PST | Palm 2
Show Abstract +
The proposed learning-based ROI-insensitive absorption coefficient (AC) extraction method in ultrasound (US) aims to improve the reliability of fatty liver diagnosis by addressing operator dependency issues. Traditional methods suffer from variability due to the selection of the region of interest (ROI) and inclusion of unwanted frames. The new approach uses feature maps and a multi-frame collaborative transformation to reduce locational and temporal dependencies. Clinical tests (N=48) showed the method's AC values strongly correlated with proton density fat fraction from MRI (R=0.94, p<0.001), improving correlation and variation by 4% and 90% respectively, indicating its potential as a new diagnostic standard.
13412-11
18 February 2025 • 4:10 PM - 4:30 PM PST | Palm 2
Show Abstract +
Hide Abstract –
3D Ultrasound tomography enables the reconstruction of quantitative attenuation volumes, however mostly neglecting the frequency dependent aspect. This paper introduces a method to reconstruct the parameters alpha_0 and y of the power absorption law. Attenuation factors for several sub-bands are detected on signal level with an FFT and STFT based methods. Volumes are then reconstructed with a ray based reconstruction per frequency sub-band. To derive alpha_0 and y, a voxel based fit is applied. The validation with simulated data shows good agreement with the ground truth by having an RMSE of 0.08 dB/cm/MHz for alpha_0 and 0.06 for y. The influence of reflection losses was tested with k-Wave simulations: an increased contrast in acoustic impedance leads to artifacts in alpha_0 and y images. First reasonable results with experimental data of a gelatine and PVC phantom are presented. The new method provides a diagnostic tool since the reconstruction of the absorption coefficient alpha_0 and the exponent y carry information which is tissue dependent.
13412-12
18 February 2025 • 4:30 PM - 4:50 PM PST | Palm 2
Show Abstract +
Tissue attenuation is generally modelled as a power law in frequency. The exact Kramers-Kronig (K-K) relation has been derived for power law frequency variation using infinite series representations of operators in certain function spaces and little known infinite series expansions of functions. We give a simple derivation using only contour integrals and show some associated results from bovine femur and phantoms. The method is extended to power laws higher than unity using a subtraction technique which is required for applications. The exact variation of the speed vs frequency is important to determine the frequency independent part of the pseudo wave number that results from a standard transformation of the variable density wave equation to the Helmholtz equation. This frequency independent part may be helpful in the estimation of density.
This simple derivation is similar to standard methods in wave analysis to derive Green’s functions and integral representations and indeed the K-K integral relation itself. The extension to a>1 is required to be physically relevant and is also carried out very simply. The singular case at a=1 is established as a limiting case.
13412-13
18 February 2025 • 4:50 PM - 5:10 PM PST | Palm 2
Show Abstract +
With fewer electrical connections and larger apertures, row-column arrays can be a potential alternative to matrix arrays in high-resolution volumetric imaging. However, these probes have a rectangular field of view (FOV), which is equal to their footprint. It is possible to expand FOV with a diverging lens, but the lens causes wave attenuation, which reduces the signal-to-noise ratio (SNR) and probe penetration depth. This paper aims to demonstrate that a large FOV can be achieved in 3D ultrasound imaging using a lens and coded excitation (CE) without negatively impacting the SNR and penetration depth. The proposed method was evaluated using a chirp signal with a center frequency of 6 MHz, a bandwidth of 4 MHz, and a duration of 5 µs. With the CE signal, the SNR increased by approximately 10 dB compared to sinusoidal excitation.
13412-14
18 February 2025 • 5:10 PM - 5:30 PM PST | Palm 2
Show Abstract +
This study presents the first experimental validation of ray-based image reconstruction techniques for high-resolution, quantitative sound speed imaging using transmission ultrasound. These methods model singly-scattered waves and reconstruct images by sequentially minimizing an objective function across increasing frequency bands, balancing computational efficiency with resolution. The methods were applied to in-vitro and in-vivo datasets from the University of Rochester Medical Center using an open-source MATLAB package. In-vitro accuracy was assessed by comparing reconstructed images with phantom data, demonstrating strong agreement. In-vivo results were evaluated against those from a full-field frequency-domain Helmholtz solver. The study highlights the computational efficiency and scalability of these techniques, demonstrating their potential for clinical ultrasound imaging. The open-source nature of these methods fosters future advancements in quantitative imaging.
Wednesday Morning Keynotes
19 February 2025 • 8:30 AM - 10:00 AM PST | Town & Country B/C
Session Chairs:
Maryam E. Rettmann, Mayo Clinic (United States), John E. Tomaszewski, Univ. at Buffalo (United States)
View Full Details: spie.org/wednesday-morning-keynotes
8:30 AM - 8:35 AM:
Welcome and introduction
8:35 AM - 8:40 AM:
Award announcements
- Robert F. Wagner Award finalists for conferences 13408 and 13413
- Early-Career Investigator Award: Image-Guided Procedures, Robotic Interventions, and Modeling
- Student Paper Award: Image-Guided Procedures, Robotic Interventions, and Modeling
13408-506
Image-guided surgery and examinations using ultrasound
(Keynote Presentation)
19 February 2025 • 8:40 AM - 9:20 AM PST | Town & Country B/C
Show Abstract +
Many of today’s cancer surgeries are carried out with robot assistance. Using real-time intra-operative ultrasound, we can overlay pre-operative imaging into the surgeon’s console, enabling visualization of sub-surface anatomy and cancer at the same time with the standard laparoscopic camera view. We will discuss aspects of system design, visualization and registration methods that enable such visualization, and present our results. We will also present tissue and instrument tracking approaches that can be used in future augmented reality systems.
For remote and underserved communities, we developed a teleultrasound approach that relies upon using a novice – the patient, a family member or friend – as a robot to carry out the examination. The novice wears a mixed reality headset and follows a rendered virtual ultrasound transducer with the actual transducer. The virtual transducer is controlled by an expert, who sees the remote ultrasound images and feels the transducer forces. This tightly-coupled expert-novice approach has advantages relative to both conventional and robotic teleultrasound. We discuss our system implementation and results.
13413-507
The future of diagnostics: the role of computational pathology in tomorrow's medicine
(Keynote Presentation)
19 February 2025 • 9:20 AM - 10:00 AM PST | Town & Country B/C
Show Abstract +
Computational Pathology has already led to remarkable innovations in diagnostics, achieving expert pathologist performance in tasks such as prostate cancer grading and cancer metastasis detection. In recent years, we have seen rapid advances, with weakly supervised models able to predict patient outcomes or genetic mutations and foundation models enabling application to rarer diseases. However, this only scratches the surface of what will be possible in the near future.
In this talk, I will briefly touch on the history of computational pathology and how we got to where we are today. Subsequently, I will highlight the current methodological innovations in the field and their potential for causing a paradigm shift in diagnostic pathology. I will discuss how these innovations, combined with the AI-driven integration of radiology, pathology, and 'omics data streams, could change the future of diagnostics as a whole. Last, I will discuss the challenges and pitfalls moving forward and how we, as a community, can contribute to addressing them.
Break
Coffee Break 10:00 AM - 10:30 AM
Session 4:
Photoacoustic Imaging
19 February 2025 • 10:30 AM - 12:10 PM PST | Palm 2
13412-15
19 February 2025 • 10:30 AM - 10:50 AM PST | Palm 2
Show Abstract +
In photoacoustic computed tomography (PACT), accurate estimation of the initial pressure (IP) distribution generally requires precise knowledge of the object's speed-of-sound (SOS) distribution, which is generally unknown. Joint reconstruction (JR) of the IP and SOS distributions from PACT measurement data offers a promising solution to this challenge by concurrently estimating both quantities. However, this joint estimation problem is ill-posed and is formulated as a non-convex optimization problem. This work investigates the effectiveness of using object constraints in improving the accuracy and stability of JR solutions in PACT. The focus is on canonical constraints — support, bound, and total variation constraints—which are grounded in prior knowledge about the properties of the imaged object. These constraints are aimed at constraining the solution space and potentially mitigating the non-uniqueness of the problem. Although these constraints are practical and straightforward to implement, their impact has not been thoroughly investigated in the context of JR for PACT. This study systematically evaluates their effectiveness through two-dimensional computer simulations using ana
13412-16
19 February 2025 • 10:50 AM - 11:10 AM PST | Palm 2
Show Abstract +
Dynamic contrast-enhanced multispectral optoacoustic tomography (DCE-MSOT) is a non-invasive imaging modality that uses multispectral optical excitation along with tomographic acoustic detection principles to estimate tumor perfusion rates in biological tissues. The reconstructed image can be viewed as a function of space, wavelength, and time. Previous works on dynamic optoacoustic imaging have demonstrated that a low-rank-based representation of the sought-after spatiotemporal object can achieve accurate, memory-efficient reconstruction of high-resolution dynamic images. This work proposes extending low-rank approaches to multispectral dynamic image reconstruction. The proposed approach is validated in a virtual imaging study and demonstrated in-vivo for DCE-MSOT imaging of tumor perfusion.
13412-17
19 February 2025 • 11:10 AM - 11:30 AM PST | Palm 2
Show Abstract +
Spectroscopic photoacoustic (sPA) imaging identifies biological materials by revealing their unique optical absorption spectra, but it is prone to noise from various sources, making denoising challenging. Traditional methods either rely on data-driven approaches, which depend on extensive training, or analytical methods, which require domain-specific knowledge. To address these limitations, we propose Spectroscopic-Zero-Shot Hybrid, a tuning-free, data-free denoising method that preserves spectral information. By integrating Zero-Shot N2N and Spectral BM3D, our approach offers stable, predictable denoising performance. Simulations and ex vivo studies validate the method's effectiveness in preserving spectrum information during denoising.
13412-19
19 February 2025 • 11:30 AM - 11:50 AM PST | Palm 2
Show Abstract +
CAR-T cell immunotherapy is a promising technique for cancer treatment. To better understand and improve its efficacy for solid tumours, methods for quantifying the CAR-T cell distribution are necessary. One approach involves inserting a reporter gene into the CAR-T cells, causing them to express photochromic proteins that provide strong near-infrared (NIR) optical contrast. NIR photoacoustic (PA) imaging is then used to image these proteins, and implicitly the CAR-T cells.
In this study machine learning techniques are used to classify and predict the spatial concentration of the photochromic proteins by analysing time series PA images. To address the need for large training datasets, we developed a novel 3D simulation framework, which generates accurately labelled PA images of CAR-T cells expressing the reporter gene. Neural networks, including a Multi-Layer Perceptron and U-Net, demonstrated superior performance over previous methods, achieving high accuracy in protein concentration prediction.
13412-20
19 February 2025 • 11:50 AM - 12:10 PM PST | Palm 2
Show Abstract +
Adoptive cell therapy (ACT) continues to fail at treating a majority of cancers. Advanced immunoimaging tools are needed to guide and enhance cellular immunotherapies to achieve greater treatment success. We developed a combined ultrasound (US) and photoacoustic (PA) imaging approach to: 1) improve T cell delivery to the primary tumor via magnetic guidance; 2) enable real-time, longitudinal monitoring of cell accumulation; and 3) assess response to therapy. Our approach leverages nanotechnology and cell surface engineering to safely tag the T cells, generate PA contrast for imaging post-transfer, and enable magnetic delivery. In vivo US/PA imaging results detected NP-labeled T cell accumulation at the tumor, visualized changes in tumor volume, and conveyed accompanying changes in blood biomarkers. Different US/PA imaging trends were observed according to a positive or negative anti-tumor response. Results demonstrate the potential of the approach and motivate further development of advanced, theranostic immunoimaging platforms.
Break
Lunch Break 12:10 PM - 1:40 PM
Session 5:
Blood Flow Imaging
19 February 2025 • 1:40 PM - 3:00 PM PST | Palm 2
13412-21
19 February 2025 • 1:40 PM - 2:00 PM PST | Palm 2
Show Abstract +
Here, a complex gradient optimization approach for precise estimation and correction of tissue motion in ultrasound data is introduced. The concept is to use gradient descent optimization to minimize the magnitude of the complex difference between neighboring frames in an ultrasound sequence. This method outperforms transverse-oscillation, speckle tracking, intensity registration-based methods for precise motion estimation in both in vivo and simulated ultrasound datasets, with sub-micron motion estimation accuracy shown in simulated data. Used with echo-cancellation in vivo for ultrasound rat data, peak tissue amplitude was reduced by a further 11 dB compared to prior motion estimation and correction methods.
13412-22
19 February 2025 • 2:00 PM - 2:20 PM PST | Palm 2
Show Abstract +
In conventional ultrasound (US) vector flow imaging (VFI) with linear array transducers, the limited aperture size restricts the detection of lateral Doppler frequency shifts, thereby limiting omnidirectional flow imaging capabilities and causing inaccuracies in vector flow reconstruction. In this study, we propose employing a ring-array US transducer that fully encloses the imaged object to interrogate frequency shifts from all directions. This approach, termed ultrasound tomographic vector flow imaging (UST-VFI), enables accurate estimation of both axial and lateral components of flow vectors based on the detected Doppler frequency shifts in each tomographic view, using a least-squares estimator. Proof-of-concept results from a rotary disk phantom demonstrated clear enhancements in uniformity and a reduction in angle-dependent errors in both the magnitude and phase of the estimated flow vectors. Once the method is fully developed, ultrasound tomography could enable the analysis of blood flow dynamics in response to pathological changes, such as breast tumor progression and treatment response.
13412-23
19 February 2025 • 2:20 PM - 2:40 PM PST | Palm 2
Show Abstract +
Tendon overuse injuries, marked by neovessel ingrowth, pose a clinical challenge. The limited observation time of conventional ultrasound imaging methods, such as color Doppler and power Doppler imaging, makes it difficult to visualize slow-flow events. These methods can only detect blood flow when the tendon is significantly diseased. A fast contrast-free super-resolution ultrasound technique (SURE) has been hypothesized to reliably reveal microvasculature in injured human patellar tendons at a resolution below λ/2. Unlike conventional Doppler, SURE identified more detailed microvasculature, achieving vessels as small as 44 μm. Consistent detection of microvascular structures was demonstrated in six 4.8-second segments, with a maximum standard deviation in vessel width of 7.0 μm.
13412-24
19 February 2025 • 2:40 PM - 3:00 PM PST | Palm 2
Show Abstract +
Thumb osteoarthritis is a prevalent degenerative joint disease. Inflammation is a key factor and contributes to vascular changes in the joint. The vascular changes and reorganization in the joint lining are not fully understood. Three-dimensional ultrasound imaging provides a method of visualizing soft tissue structures of the joint affected by inflammation and can detect blood flow with Doppler technologies. The three-dimensional ultrasound imaging system provides volumetric analysis of the joint, including measures of joint blood flow. This work investigated the test-retest reliability of the three-dimensional ultrasound Doppler blood flow measure in thumb osteoarthritis patients. The development of these volumetric ultrasound imaging tools and measures for musculoskeletal imaging can further the understanding of inflammation and blood flow changes in the disease.
Break
Coffee Break 3:00 PM - 3:30 PM
Session 6:
Ultrasound Computed Tomography
19 February 2025 • 3:30 PM - 5:10 PM PST | Palm 2
13412-25
19 February 2025 • 3:30 PM - 3:50 PM PST | Palm 2
Show Abstract +
Ultrasound tomography (USCT) in combination with advanced reconstruction algorithms such as full waveform
inversion (FWI) leads to ultrasound images with an image quality comparable to that of MRI images. The aim of
the QUSTom project is to extend the application of such algorithms to full 3D imaging. A key component of the
project is a clinical feasibility study evaluating the performance of the 3D FWI method. This paper focuses on
the technical aspects of data acquisition from corresponding datasets using the 3DUSCTIII device as part of this
study.
13412-26
19 February 2025 • 3:50 PM - 4:10 PM PST | Palm 2
Show Abstract +
3D ultrasound tomography is now deployed in multiple breast clinics in 5 countries. The results here and in development show image quality comparable to MRI, but it is also quantitative. This never before published work quantifies the exact ratio of the data contribution to a single voxel and the relative computational complexity in the 3D ultrasound case vs MRI.
Even with the newer model based methods of MRI reconstruction the data acquisition (DA) is such that there are 5+ orders of magnitude less data contributing to a voxel for MRI than for quantitative 3D ultrasound due to the 3D nature of the acoustic field.
However, this also requires 8+ orders of magnitude more calculations than MRI and the solution to this problem is reviewed. This explains the quantitative accuracy, high resolution and remarkable capabilities of 3D ultrasound in the presence of bone and air. Images from orthopedics, breast and whole body imaging indicate image quality. Ewald circle analysis indicates theoretical resolution and is compared with measured data.
The observed quantitative accuracy, high resolution and repeatability even in the presence of bone and air can be partially explained in this way.
13412-27
19 February 2025 • 4:10 PM - 4:30 PM PST | Palm 2
Show Abstract +
Ultrasound computed tomography (USCT) aims to reconstruct high-resolution maps of acoustic tissue properties, such as speed-of-sound (SOS). Previous work proposed a learned reconstruction method using paired traveltime (TT) and reflection tomography (RT) images as inputs for efficient high-resolution SOS estimation. This study extends the previous virtual imaging work to real-world validation using clinical data. A convolutional neural network was trained on clinical data, using SOS maps from full-waveform inversion as targets. The experiments demonstrated the dual-input-modality IILR method's effectiveness at providing high-resolution and accurate SOS estimates, with a dramatically reduced computational burden compared to FWI. The method also demonstrated the ability to mitigate certain artifacts that were present in the FWI target images. This work suggests the promise of the learned reconstruction method for enabling clinical applications of breast USCT in low resource settings.
13412-28
19 February 2025 • 4:30 PM - 4:50 PM PST | Palm 2
Show Abstract +
This work applies optimal transport within transcranial ultrasound imaging and demonstrates the efficacy of this misfit function on a dataset collected for a tissue mimicking skull phantom. Transcranial ultrasound imaging approaches using full-waveform inversion have historically suffered from severe cycle skipping effects, particularly when starting the inversion from a homogeneous initial model. In order to overcome cycle skipping for this application, an optimal transport misfit function is used to reduce the non-convexity of the misfit landscape while significantly improving the prominence of relevant features within the gradients computed using the adjoint method. This misfit function, which accounts for mismatches in both amplitude and phase, is very well-suited for transcranial applications where the skull is absent from the initial model. This approach is demonstrated on a set of laboratory data collected for a tissue mimicking skull phantom.
13412-29
19 February 2025 • 4:50 PM - 5:10 PM PST | Palm 2
Show Abstract +
3D ultrasound tomography is already in use in clinics for breast. This is a proof of concept showing evidence that transcranial acoustic propagation is possible in neonates and shows actual reconstruction of excised ovine brain and excised bovine eye to establish high resolution images of transcranial/near cranial organs relevant to neonatology. Hemorrhage in newborn brain is a critical medical condition. MRI is often used but may not be available in low resource environments (LRE). 3D ultrasound does not require radiation or Gadolinium. MRI may require dangerous sedation. We show images of speed of sound (SOS) and reflection of the bovine eye and reflection images (corrected for SOS and refraction) that indicate proof of concept. We quantify resolution for reflection at 0.8 mm. The eye images are superior to MRI in resolution and quantitative accuracy, and the brain images show comparable resolution and detail to MRI. The reflection images use the high resolution SOS image for refraction and SOS correction, which increases resolution and removes speckle, leading to MRI-like images useful for hemorrhagic diagnosis in neonates without dangerous radiation or Gadolinium contrast.
Posters - Wednesday
19 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
View Full Details: spie.org/mi/wednesday-poster-session
Conference attendees are invited to attend the SPIE Medical Imaging poster session on Wednesday evening. Come view the posters, enjoy light refreshments, ask questions, and network with colleagues in your field. Authors of poster papers will be present to answer questions concerning their papers. Attendees are required to wear their conference registration badges.
Poster Presenters:
Poster Setup and Pre-Session Viewing: 7:30 AM - 5:30 PM Wednesday
- In order to be considered for a poster award, it is recommended to have your poster set up by 1:00 PM Wednesday. Judging may begin after this time. Posters must remain on display until the end of the Wednesday evening poster session but may be left hanging until 10:00 AM Thursday. After 10:00 AM, any posters left hanging will be discarded.
spie.org/MI/Poster-Presentation-Guidelines
13412-36
19 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
We demonstrate passive ultrasound imaging algorithms that do not use time-of-flight information. Using a two-element array of biaxial ultrasound sensors, where biaxial sensors are distinguished by their ability to independently detect the direction of arrival (DOA) of incident ultrasound, we demonstrate biaxial image reconstruction algorithms that use either one or two sensors to map acoustic energy in one- or two-dimensions, respectively. To the best of our knowledge, this is the first demonstration of passive acoustic mapping using a single sensor as well as the first demonstration of two-dimensional passive acoustic mapping using only DOA information. These results are contrasted with conventional TOF-based methods, which fail with so few sensors. These results enable smaller, more compact arrays for applications where space comes at a premium.
13412-37
19 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Stomach cancer is the fifth most common cancer and the third leading cause of cancer-related deaths globally. Early detection is crucial for improving survival rates and treatment outcomes. Current imaging methods like CT, MRI, and endoscopic ultrasound (EUS) have limitations such as low sensitivity for early tumours, ionizing radiation, high costs, and invasiveness, which limit their use in broad population screenings. Ultrasound imaging could address these challenges, but conventional methods lack sufficient image quality for stomach imaging. This study investigates the potential of full-waveform inversion (FWI), an advanced imaging algorithm from seismology, for improving ultrasound stomach imaging. Simulation results with a simplified numerical abdomen phantom show promising potential.
13412-38
19 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Ultrasound computed tomography (UCT) has emerged as a promising modality to produce high-resolution and isotropic anatomical visuals. Yet, collecting extensive UCT data via numerous transmissions is time-intensive. Sparse transmission strategies offer a practical solution to streamline data collection, but traditional Delay-and-Sum (DAS) approaches can significantly compromise image quality. Addressing these challenges, this research introduces an efficient framework for the reconstruction of reflection UCT images utilizing sparse transmission data, grounded in a novel residual-based diffusion probabilistic model. This method employs a Markov chain to enable seamless transitions between high and low-resolution images by manipulating residuals. Through evaluation experiments utilizing in-vivo human limb imaging data, we demonstrate the ability of our proposed method to produce high-quality reflection UCT images with a reduced transmission count. Quantitative analysis demonstrates the method's proficiency in reconstructing superior reflection UCT images from limited transmission data.
13412-39
19 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Levator ani muscle (LAM) avulsion is a pelvic floor injury often resulting from vaginal childbirth and is linked to the development of pelvic organ prolapse. A previous publication has indicated that up to 40% of women worldwide will experience this pelvic floor disorder in their lifetime. Diagnosis of LAM avulsion is commonly done through ultrasound imaging and requires trained experts, often leading to weeks-long delays in receiving diagnostic results and treatment. This study developed a deep learning system to automatically classify the degree of LAM avulsion from 3D transperineal ultrasound images, aiming to expedite the diagnostic process. By analyzing these images, our model accurately identifies the presence and severity of LAM avulsion in each patient. This automated approach addresses the challenges of manual ultrasound diagnosis and enhances screening accessibility, benefiting areas with limited healthcare resources, by reducing the need and time for expert review.
13412-41
19 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Osteoarthritis (OA) is the most common form of arthritis and is characterized by inflammation of the synovium which plays a role in pain production and cartilage degradation. To visualize and quantify the inflammation, our lab has developed a 3D US musculoskeletal imaging system, equipped with a mechatronic counterbalanced arm that supports a 3D US-based linear motorized scanner and allows tracking of the 3D US scanner. The work being presented describes the validation of the system’s tracking and image fusion properties, as well as a demonstration of the system’s utility for imaging large joints, such as the knee. The validation was conducted using agar phantoms containing inclusions for spatial and volumetric validation. Results indicated accurate and precise tracking and fusion, making the system a promising tool for quantitative monitoring of soft-tissue changes in joints afflicted with musculoskeletal pathologies.
13412-42
19 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Ultrasound is used as a supplementary diagnostic tool to diagnose breast cancer after initial screening mammography, due to its high false positive rate. To improve upon the false positive rate, computer aided diagnostic systems have been introduced that analyze grayscale images. However, the conversion to grayscale results in a significant loss of acoustic information. Therefore, this study explores the novel fusion of seven different raw quantitative ultrasound features extracted from the raw data to classify breast ultrasound images. With these added features, our algorithm shows improved accuracy by 15.62% compared to grayscale alone, which is promising to improve breast ultrasound.
13412-43
19 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Ultrasound can be used to stimulate the mouse brain, but due to aberrations caused by the skull, it is difficult to focus ultrasound energy where it is desired. For this reason, acoustic hologram lenses have been shown to focus effectively by manipulating the phase. However, since hologram lenses are generated by considering the position of the skull, accurate stimulation is difficult if the position of the ultrasound transducer and the mouse skull is not correct. For this reason, we checked the focusing of the ultrasound depending on the position and angle of the transducer with the hologram lens and the mouse skull. This study provides a guideline for the point at which the performance degrades significantly if the position or angle deviates beyond a certain point.
13412-44
19 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
High-quality datasets are essential for effective deep learning training. However, collecting consistent ultrasound image data is challenging due to privacy concerns, patients' physical characteristics, and equipment variability. To address this, we propose a CycleGAN-based method to transform real ultrasound data to resemble data generated by the Field II simulation program. This method creates synthetic data that preserves the image structure of real data while incorporating the detailed characteristics of the simulation, facilitating the model's application to real datasets. By bridging the domain gap, the synthetic data enhances feature learning from both real and simulated datasets. Consequently, experiments using synthetic data show higher performance compared to those using only real data or data augmented with simulation alone.
13412-45
19 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Topological data/image analysis has grown into an independent area of practical research in the past 8 years due to the increased availability of software to carry out persistent homology, etc. Other aspects include cohomology, graph theory, Morse theory and homotopy groups. We focus here on homotopy groups. We introduce a specific map from S1 to the N-torus, where N is the number of elements in the receiver array in the QT Imaging system. We show the homotopic equivalence classes of these maps (the elements of the (known) first homotopy group of the N-Torus) have direct relevance to phase wrapping in frequency domain 3D ultrasound tomography in a manner somewhat different from the normal understanding of phase wrapping/cycle-skipping.
The maps are related to the Lie symmetry group associated with full view tomography: SO(2)xR which arise in the equivalence of our method to training a neural net previously shown. This characterization as a known homotopy group gives constraints on frequency stepping in reconstruction and/or vertical differences between data levels in data acquisition scenarios, depending on the interpretation of the R coordinate in the Lie symmetry group.
13412-46
19 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Cerebral palsy (CP) is a neurodevelopmental disorder that can have a negative impact on the tendons and muscles of developing children. This study aims to understand the difference in morphological properties of the patellar tendons of children with CP and compare them with tendons of typically developing (TD) children to guide physical therapy treatment. Twenty children were recruited and compared, leading to statistically significant differences in Rayleigh and Nakagami distribution statistics, σ, Ω, and m in age and gender-matched children. The results found are promising for the continued investigation of using quantitative ultrasound for future guidance in physical therapy.
13412-47
19 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Multi-element ultrasound arrays are not always perfect, especially when produced using experimental technology.
Element variability leads to deterioration of image quality, which prevents a fair comparison to more established
transducer manufacturing technologies. This paper presents an approach to compensate for the geometric (Z-
axis) imperfections of a transducer, relying on the characterization results of its individual elements. It is
hypothesized that compensating for the geometric imperfection of the transducer can significantly improve the
ultrasound image quality. The potential of the approach is evaluated through image quality comparison using a
commercial transducer, an experimental transducer, and a multi-wire phantom. For a wire at 75 mm depth, the
lateral width of a wire image is reduced from 7.4 λ to 3.1 λ, and the cystic contrast (r = 2.5λ) is improved from
-21.4 dB to -31.1 dB.
13412-48
19 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Surgical resection is vital to the treatment workflow for oral cavity squamous cell carcinoma. These tumors are aggressive and fast-growing, meaning they may change in size and shape from the time of biopsy/diagnosis and surgery, approximately 1-2 months later. Thus, when the surgeon begins to resect the malignant tissue, they are often faced with a more invasive tumor volume than originally indicated in the report. We aim to use inexpensive, bedside point-of-care imaging to quantify the tumor volume just prior to surgery to provide the clinician with accurate, updated soft tissue information about the diseased area.
13412-49
19 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Clipping artifacts impact the image quality and diagnostic accuracy in ultrasound imaging. This study introduces a post-processing technique designed to eliminate clipping in ultrasound scans acquired using synthetic aperture ultrasound systems. It is hypothesized that the proposed method can effectively mask out clipped signals while preserving non-clipped contributions and without a significant loss of image quality.
A phantom and an in-vivo rat brain scan were used to test this hypothesis. The analysis of the phantom data showed that the proposed method can be applied to remove mild clipping levels without a significant loss in contrast and resolution. In fact, when up to 30\% of the data was clipped, the loss in resolution after applying clipping removal was lower than 2.5\% and the contrast decreased less than 7.9\% when compared with the original non-clipped image.
In-vivo results, on the other hand, demonstrated that applying the proposed algorithm effectively removes clipping artifacts while preserving the underlying information. Additionally, it enhances the image's dynamic range by approximately 12\%, when compared with the clipped data.
13412-50
19 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
The Left Atrial Appendage (LAA) is a complex anatomical structure often responsible for cardiac thrombi. Transesophageal echocardiography (TEE) is the primary imaging modality for detailed LAA assessment and thrombus detection. Segmenting the LAA is crucial, yet existing 3D TEE-based methods often require manual centerline initialization, making them semi-automatic. This study aimed to automate the LAA centerline stage using deep learning (DL) techniques. A UNET-based DL model was trained and validated on 188 3D TEE volumes, with manual centerline annotations serving as the ground truth. Following automated centerline identification, state-of-the-art segmentation was applied, and the results were compared to manually corrected segmentations. The automated method achieved high segmentation performance, with an Average Distance Error of 0.60±0.46mm, a Dice Coefficient of 94.45±4.25%, and a 95% Hausdorff Distance of 2.21±1.86mm. These findings demonstrate the potential of DL methods to fully automate LAA segmentation in 3D TEE images.
13412-51
19 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Noninvasive imaging of deep-tissue microvascular structures is essential for accurate clinical diagnosis and monitoring. Ultrasound Localization Microscopy (ULM) achieves subwavelength resolution imaging but is hindered by challenges such as extended acquisition times, high microbubble (MB) concentration requirements, and localization inaccuracies. In this study, we propose ENS-ULM, an ensemble model that integrates a Swin transformer with a subpixel Convolutional Neural Network (CNN) to enhance MB localization in ULM. Using synthetic data, we validated ENS-ULM with metrics including the Jaccard index and localization precision. Our ensemble model outperformed traditional approaches, such as Gaussian Fitting and Radial Symmetry, as well as individual CNN and Swin transformer models, achieving superior imaging precision and accuracy. These findings highlight the potential of ensemble methods in advancing MB localization performance for ULM applications.
13412-52
19 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
Diffusion models for image generation have been a subject of increasing interest due to their ability to generate diverse, high-quality images. Image generation has immense potential in medical imaging because open-source medical images are difficult to obtain compared to natural images, especially for rare conditions. The generated images can be used later to train classification and segmentation models. In this paper, we propose simulating realistic ultrasound (US) images by successive fine-tuning of large diffusion models on different publicly available databases. To do so, we fine-tuned Stable Diffusion, a state-of-the-art latent diffusion model, on BUSI (Breast US Images) an ultrasound breast image dataset. We successfully generated high-quality US images of the breast using simple prompts that specify the organ and pathology, which appeared realistic to three experienced US scientists and a US radiologist. Additionally, we provided user control by conditioning the model with segmentations through ControlNet. We will release the source code at http://code.sonography.ai/ to allow fast US image generation to the scientific community.
13412-53
19 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
BreastLightSAM is a lightweight pipeline developed for rapid and precise breast cancer diagnosis utilizing 2D ultrasound images. The model incorporates an optimized Segmentation Anything Model (SAM) with a RepViT-based encoder for segmentation, and a compact classification module. This integration yields high accuracy, achieving a mean Dice Similarity Coefficient (DSC) of 0.812 and a mean pixel accuracy of 0.943, with minimal latency of just 28.6 milliseconds, rendering it suitable for mobile and resource-constrained environments. Through the incorporation of human-like prompt design, BreastLightSAM enhances robustness and generalization, demonstrating significant potential for real-time clinical applications.
13412-54
19 February 2025 • 5:30 PM - 7:00 PM PST | Golden State Ballroom
Show Abstract +
This study integrates radiomics with ultrasound (US) radiofrequency (RF) data to enhance stromal nodule detection in ex-vivo prostate specimens with benign prostatic hyperplasia (BPH). By combining RF data and B-mode images, the regions of interest in the prostate were analyzed. Annotated RGB images identified stromal and non-stromal regions, which were correlated with US data. Radiomic features, including first-order energy and GLCM-based metrics, were extracted using PyRadiomics. Machine learning models showed that RF-derived radiomic features significantly improve stromal nodule detection, offering a non-invasive, precise diagnostic approach.The integration of radiomics with RF data offers a non-invasive, precise method for distinguishing between stromal and non-stromal regions of prostate, promising improved patient outcomes and personalized treatment plans.
Thursday Morning Keynotes
20 February 2025 • 8:30 AM - 10:00 AM PST | Town & Country B/C
Session Chairs:
Susan M. Astley, The Univ. of Manchester (United Kingdom), Andrzej Krol, SUNY Upstate Medical Univ. (United States)
View Full Details: spie.org/thursday-morning-keynotes
8:30 AM - 8:35 AM:
Welcome and introduction
8:35 AM - 8:40 AM:
Award announcements
- Robert F. Wagner Award finalists for conferences 13407 and 13410
- Computer-Aided Diagnosis Best Paper Award
13407-508
Pioneering vision: The journey of building AI for transformative healthcare
(Keynote Presentation)
20 February 2025 • 8:40 AM - 9:20 AM PST | Town & Country B/C
Show Abstract +
Medical imaging has long been a driving force in healthcare innovation, paving the way for digital transformation. In this session, we’ll explore the evolution of AI in medical imaging - where it started, the challenges of real-world adoption, and the breakthroughs shaping its future. From early academic research to real-world clinical integration, we'll examine how AI is moving beyond theoretical potential to deliver real impact in patient care. Join us for an engaging discussion on how engineering, data science, and AI are redefining healthcare workflows, and what's next for those leading the charge in this transformation.
13410-509
Machine learning in neuroimaging: Understanding heterogeneity of neurologic diseases and building personalized imaging-AI biomarkers
(Keynote Presentation)
20 February 2025 • 9:20 AM - 10:00 AM PST | Town & Country B/C
Show Abstract +
Machine learning has transformed medical imaging in general, and neuroimaging in particular, during the past two decades. We review our work in this field, starting with early contributions on developing personalized predictive markers of brain change in aging and Alzheimer’s Disease, and moving to recent weakly-supervised deep learning methods, aiming to dissect heterogeneity of brain change in neurodegenerative and neuropsychiatric diseases, as well as in brain cancer. We show that disease-related brain changes can follow multiple trajectories and patterns, which have distinct clinical and genetic correlates, thereby suggesting a dimensional approach to capturing brain phenotypes, using machine learning methods.
Break
Coffee Break 10:00 AM - 10:30 AM
Session 7:
Applications of Machine Learning in Ultrasound Imaging
20 February 2025 • 10:30 AM - 12:30 PM PST | Palm 2
13412-30
20 February 2025 • 10:30 AM - 10:50 AM PST | Palm 2
Show Abstract +
Compared to conventional two-dimensional (2D) ultrasound, three-dimensional (3D) ultrasound (US) images are a more sensitive alternative for monitoring size and shape of neonatal cerebral lateral ventricles when diagnosing intraventricular hemorrhaging (IVH). The ventricles must be segmented by an expert to estimate the ventricular volume which can be time-consuming and difficult to obtain. In this paper, we describe a scribble-based weakly supervised segmentation method that trains only on non-expert drawn scribbles. We trained and tested two segmentation methods, a vanilla 3D U-Net benchmark and a weakly supervised learning for medical image segmentation (WSL4MIS) method built into the 3D U-Net model, using 56 3D US images. For all experiments, the WSL4MIS method had a higher mean DSC and lower standard deviation than that of the baseline 3D U-Net when both used scribble data for training.
13412-31
20 February 2025 • 10:50 AM - 11:10 AM PST | Palm 2
Show Abstract +
The QTI Breast Acoustic CTTM scanner currently uses 2-mm vertical spacing between measurement levels to create high-quality images. This work aims to reduce the number of measurements by increasing the spacing to 4-mm and using a deep learning model to reconstruct the missing levels, restoring the original 2-mm spacing. The method uses adaptive convolution inspired by video frame interpolation, combined with a conditional Generative Adversarial Network where the generator fills in missing data and the discriminator checks its realism. The approach also includes multi-scale learning and residual learning in the deep learning model as well as wavelet denoising in preprocessing to smooth signals, remove noise and capture fine details. This approach can cut patient scan time almost in half, doubling the system's operational efficiency while maintaining image quality, offering a practical solution for faster clinical imaging while maintaining accuracy.
13412-32
Mask-enhanced deep-learning for prostate ultrasound tomography with narrow data acquisition aperture
20 February 2025 • 11:10 AM - 11:30 AM PST | Palm 2
Show Abstract +
Prostate cancer is a leading cause of cancer-related deaths. Traditional ultrasound methods struggle with sensitivity and specificity due to narrow data acquisition apertures. This study uses a convolutional neural network (CNN) to enhance prostate imaging. Validated with synthetic phantoms and finite difference simulations, our CNN-based approach surpasses traditional full-waveform inversion (FWI) in accuracy and efficiency, swiftly reconstructing high-resolution speed of sound (SOS) maps. This method promises significant improvements in prostate cancer diagnosis and treatment planning, reducing the need for invasive procedures and enabling better clinical decisions.
13412-33
20 February 2025 • 11:30 AM - 11:50 AM PST | Palm 2
Show Abstract +
The super-resolution (SR) of 3D ultrasound (US) images reconstructs and increases the number of slices, maintaining the high-resolution (HR) image visually like the corresponding low-resolution (LR) while improving anatomical structures' visibility and avoiding artefacts. We propose a novel learning-based super-resolution method of 3D US images that accounts for the presence of speckle noise and preserves the properties of the high-resolution image in terms of texture patterns and visual appearance. We specialise our model to speckle noise and 2X and 4X up-sampling factors. We apply the trained network to reconstruct non-acquired slices of 3D US images from the forearm anatomical district of an open US data set. Finally, we evaluate the quantitative results compared to previous work according to feature, local, and noise-based metrics.
13412-34
20 February 2025 • 11:50 AM - 12:10 PM PST | Palm 2
Show Abstract +
PURPOSE: Deep learning is promising for enabling accurate and automatic prostate segmentation. Existing deep learning segmentation model approaches often rely on large training datasets for good generalization. We aim to adapt pre-trained foundation models to enable accurate, data-efficient, and robust prostate segmentation. METHODS: We adapt the promptable foundation model SAM for prostate segmentation on rotational micro-ultrasound scans of the prostate. We design novel prompting strategies to provide the model with enhanced 3D context awareness to improve segmentation. RESULTS: Our model, which we call SliceTrack-SAM, outperforms prior state-of-the-art in micro-ultrasound prostate segmentation. We achieve a Dice coefficient of 94.0% and Hausdorff distance 1.76 mm on the MicroSegNet dataset. Qualitative analysis and ablation studies further validate the success of our approach. CONCLUSION: Transfer learning from pre-trained foundation models can alleviate the challenge of data scarcity and improve the generalization of medical imaging deep learning systems. Prompts provide a flexible and effective way to leverage auxiliary information, such as 3D context, when using these model.
13412-35
A comparative study of deep-learning methods for left atrial appendage segmentation in 3D TEE images
20 February 2025 • 12:10 PM - 12:30 PM PST | Palm 2
Show Abstract +
Atrial fibrillation (AF) is the most common cardiac arrhythmia, affecting ~2.5% of the world's population. In patients with nonvalvular AF, about 90% of thrombi are located in the left atrial appendage (LAA). LAA closure (LAAC) has emerged as an alternative, performed through percutaneous implantation of an occlusion device guided by fluoroscopy, but it relies heavily on user experience and involves high radiation exposure. Transesophageal echocardiography (TEE) is an imaging modality that offers 3D real-time visualization, radiation-free, essential for guiding the LAAC procedure. Therefore, LAA segmentation from ultrasound(US) imaging is crucial. Thus, we conducted a comparative analysis of four deep learning (DL) networks for segmenting 3D US-TEE images of the LAA: UNET, UNETR, DynUNET, and ViTAutoEnc. Using a dataset of 188 3D US-TEE image volumes, DynUNET stood out with an average distance error of 1.80±0.83mm, Dice Coefficient of 82.21±5.89%, and 95% Hausdorff Distance of 2.40±0.92mm, demonstrating the potential of DL methods for LAA segmentation.