3-D Stochastic Numerical Breast Phantoms for Enabling Virtual Imaging Trials of Ultrasound Computed Tomography

Learning a stable approximation of an existing but unknown inverse mapping: application to the half-time circular Radon transform

Supervised deep learning-based methods have inspired a new wave of image reconstruction methods that implicitly learn effective regularization strategies from a set of training data. While they hold potential for improving image quality, they have also raised concerns regarding their robustness. Instabilities can manifest when learned methods are applied to find approximate solutions to ill-posed image reconstruction problems for which a unique and stable inverse mapping does not exist, which is a typical use case. In this study, we investigate the performance of supervised deep learning-based image reconstruction in an alternate use case in which a stable inverse mapping is known to exist but is not yet analytically available in closed form. For such problems, a deep learning-based method can learn a stable approximation of the unknown inverse mapping that generalizes well to data that differ significantly from the training set. The learned approximation of the inverse mapping eliminates the need to employ an implicit (optimization-based) reconstruction method and can potentially yield insights into the unknown analytic inverse formula. The specific problem addressed is image reconstruction from a particular case of radially truncated circular Radon transform (CRT) data, referred to as 'half-time' measurement data. For the half-time image reconstruction problem, we develop and investigate a learned filtered backprojection method that employs a convolutional neural network to approximate the unknown filtering operation. We demonstrate that this method behaves stably and readily generalizes to data that differ significantly from training data. The developed method may find application to wave-based imaging modalities that include photoacoustic computed tomography.

Report on the AAPM Grand Challenge on deep generative modeling for learning medical image statistics

Background

The findings of the 2023 AAPM Grand Challenge on Deep Generative Modeling for Learning Medical Image Statistics are reported in this Special Report.

Purpose

The goal of this challenge was to promote the development of deep generative models for medical imaging and to emphasize the need for their domain-relevant assessments via the analysis of relevant image statistics.

Methods

As part of this Grand Challenge, a common training dataset and an evaluation procedure was developed for benchmarking deep generative models for medical image synthesis. To create the training dataset, an established 3D virtual breast phantom was adapted. The resulting dataset comprised about 108,000 images of size 512×512. For the evaluation of submissions to the Challenge, an ensemble of 10,000 DGM-generated images from each submission was employed. The evaluation procedure consisted of two stages. In the first stage, a preliminary check for memorization and image quality (via the Fréchet Inception Distance (FID)) was performed. Submissions that passed the first stage were then evaluated for the reproducibility of image statistics corresponding to several feature families including texture, morphology, image moments, fractal statistics and skeleton statistics. A summary measure in this feature space was employed to rank the submissions. Additional analyses of submissions was performed to assess DGM performance specific to individual feature families, the four classes in the training data, and also to identify various artifacts.

Results

Fifty-eight submissions from 12 unique users were received for this Challenge. Out of these 12 submissions, 9 submissions passed the first stage of evaluation and were eligible for ranking. The top-ranked submission employed a conditional latent diffusion model, whereas the joint runners-up employed a generative adversarial network, followed by another network for image superresolution. In general, we observed that the overall ranking of the top 9 submissions according to our evaluation method (i) did not match the FID-based ranking, and (ii) differed with respect to individual feature families. Another important finding from our additional analyses was that different DGMs demonstrated similar kinds of artifacts.

Conclusions

This Grand Challenge highlighted the need for domain-specific evaluation to further DGM design as well as deployment. It also demonstrated that the specification of a DGM may differ depending on its intended use.

A High-Resolution 3D Ultrasound Imaging System Oriented towards a Specific Application in Breast Cancer Detection Based on a 1 × 256 Ring Array

This paper presents the design and development of a high-resolution 3D ultrasound imaging system based on a 1 × 256 piezoelectric ring array, achieving an accuracy of 0.1 mm in both ascending and descending modes. The system achieves an imaging spatial resolution of approximately 0.78 mm. A 256 × 32 cylindrical sensor array and a digital phantom of breast tissue were constructed using the k-Wave toolbox. The signal is acquired layer by layer using 3D acoustic time-domain simulation, resulting in the collection of data from each of the 32 layers. The 1 × 256 ring array moves on a vertical trajectory from the chest wall to the nipple at a constant speed. A data set was collected at intervals of 1.5 mm, resulting in a total of 32 data sets. Surface rendering and volume rendering algorithms were used to reconstruct 3D ultrasound images from the volume data obtained via simulation so that the smallest simulated reconstructed lesion had a diameter of 0.3 mm. The reconstructed three-dimensional image derived from the experimental data exhibits the contour of the breast model along with its internal mass. Reconstructable dimensions can be achieved up to approximately 0.78 mm. The feasibility of applying the system to 3D breast ultrasound imaging has been demonstrated, demonstrating its attributes of resolution, precision, and exceptional efficiency.

Learned Full Waveform Inversion Incorporating Task Information for Ultrasound Computed Tomography

Ultrasound computed tomography (USCT) is an emerging imaging modality that holds great promise for breast imaging. Full-waveform inversion (FWI)-based image reconstruction methods incorporate accurate wave physics to produce high spatial resolution quantitative images of speed of sound or other acoustic properties of the breast tissues from USCT measurement data. However, the high computational cost of FWI reconstruction represents a significant burden for its widespread application in a clinical setting. The research reported here investigates the use of a convolutional neural network (CNN) to learn a mapping from USCT waveform data to speed of sound estimates. The CNN was trained using a supervised approach with a task-informed loss function aiming at preserving features of the image that are relevant to the detection of lesions. A large set of anatomically and physiologically realistic numerical breast phantoms (NBPs) and corresponding simulated USCT measurements was employed during training. Once trained, the CNN can perform real-time FWI image reconstruction from USCT waveform data. The performance of the proposed method was assessed and compared against FWI using a hold-out sample of 41 NBPs and corresponding USCT data. Accuracy was measured using relative mean square error (RMSE), structural self-similarity index measure (SSIM), and lesion detection performance (DICE score). This numerical experiment demonstrates that a supervised learning model can achieve accuracy comparable to FWI in terms of RMSE and SSIM, and better performance in terms of task performance, while significantly reducing computational time.

A Forward Model Incorporating Elevation-Focused Transducer Properties for 3-D Full-Waveform Inversion in Ultrasound Computed Tomography

Ultrasound computed tomography (USCT) is an emerging medical imaging modality that holds great promise for improving human health. Full-waveform inversion (FWI)-based image reconstruction methods account for the relevant wave physics to produce high spatial resolution images of the acoustic properties of the breast tissues. A practical USCT design employs a circular ring-array comprised of elevation-focused ultrasonic transducers, and volumetric imaging is achieved by translating the ring-array orthogonally to the imaging plane. In commonly deployed slice-by-slice (SBS) reconstruction approaches, the 3-D volume is reconstructed by stacking together 2-D images reconstructed for each position of the ring-array. A limitation of the SBS reconstruction approach is that it does not account for 3-D wave propagation physics and the focusing properties of the transducers, which can result in significant image artifacts and inaccuracies. To perform 3-D image reconstruction when elevation-focused transducers are employed, a numerical description of the focusing properties of the transducers should be included in the forward model. To address this, a 3-D computational model of an elevation-focused transducer is developed to enable 3-D FWI-based reconstruction methods to be deployed in ring-array-based USCT. The focusing is achieved by applying a spatially varying temporal delay to the ultrasound pulse (emitter mode) and recorded signal (receiver mode). The proposed numerical transducer model is quantitatively validated and employed in computer simulation studies that demonstrate its use in image reconstruction for ring-array USCT.

Stochastic three-dimensional numerical phantoms to enable computational studies in quantitative optoacoustic computed tomography of breast cancer

Significance

When developing a new quantitative optoacoustic computed tomography (OAT) system for diagnostic imaging of breast cancer, objective assessments of various system designs through human trials are infeasible due to cost and ethical concerns. In prototype stages, however, different system designs can be cost-efficiently assessed via virtual imaging trials (VITs) employing ensembles of digital breast phantoms, i.e., numerical breast phantoms (NBPs), that convey clinically relevant variability in anatomy and optoacoustic tissue properties.

Aim

The aim is to develop a framework for generating ensembles of realistic three-dimensional (3D) anatomical, functional, optical, and acoustic NBPs and numerical lesion phantoms (NLPs) for use in VITs of OAT applications in the diagnostic imaging of breast cancer.

Approach

The generation of the anatomical NBPs was accomplished by extending existing NBPs developed by the U.S. Food and Drug Administration. As these were designed for use in mammography applications, substantial modifications were made to improve blood vasculature modeling for use in OAT. The NLPs were modeled to include viable tumor cells only or a combination of viable tumor cells, necrotic core, and peripheral angiogenesis region. Realistic optoacoustic tissue properties were stochastically assigned in the NBPs and NLPs.

Results

To advance optoacoustic and optical imaging research, 84 datasets have been released; these consist of anatomical, functional, optical, and acoustic NBPs and the corresponding simulated multi-wavelength optical fluence, initial pressure, and OAT measurements. The generated NBPs were compared with clinical data with respect to the volume of breast blood vessels and spatially averaged effective optical attenuation. The usefulness of the proposed framework was demonstrated through a case study to investigate the impact of acoustic heterogeneity on OAT images of the breast.

Conclusions

The proposed framework will enhance the authenticity of virtual OAT studies and can be widely employed for the investigation and development of advanced image reconstruction and machine learning-based methods, as well as the objective evaluation and optimization of the OAT system designs.

Ultrasound computed tomography based on full waveform inversion with source directivity calibration

Ultrasound computed tomography based on full waveform inversion has the potential to provide high-resolution images of human tissues in a quantitative manner. A successful ultrasound computed tomography system requires the decent knowledge of acquisition array, including the spatial position and the directivity of each transducer, to meet the high-level demand of clinical applications. The conventional full waveform inversion algorithm assumes a point source with the omni-directional emission. Such assumption does not hold when the directivity of emitting transducer is not negligible. For a practical implementation, an efficient and accurate self-checking evaluation of directivity is crucial prior to the reconstruction of images. We propose to measure the directivity of each emitting transducer using the full-matrix captured data obtained with a water-immersed and target-free experiment. We introduce the weighted virtual point-source array to act as the proxy of emitting transducer during the numerical simulation. The weights of different points in the virtual array can be calculated from the observed data using the gradient-based local optimization method. Although the full waveform imaging method relies on the finite-difference solver of wave equation, such directivity estimation benefits from the introduction of analytical solver. The trick significantly reduces the numerical cost, enabling an automatic directivity self-check at boot. We verify the feasibility, efficiency, and accuracy of the virtual array method through simulated and experimental tests. For the experimental test, we also illustrate that full waveform inversion with directivity calibration can reduce the artifacts introduced by the conventional point source assumption, improving the quality of reconstructed images..

Dynamic Speed of Sound Adaptive Transmission-Reflection Ultrasound Computed Tomography

Ultrasound computed tomography (USCT) can visualize a target with multiple imaging contrasts, which were demonstrated individually previously. Here, to improve the imaging quality, the dynamic speed of sound (SoS) map derived from the transmission USCT will be adapted for the correction of the acoustic speed variation in the reflection USCT. The variable SoS map was firstly restored via the optimized simultaneous algebraic reconstruction technique with the time of flights selected from the transmitted ultrasonic signals. Then, the multi-stencils fast marching method was used to calculate the delay time from each element to the grids in the imaging field of view. Finally, the delay time in conventional constant-speed-assumed delay and sum (DAS) beamforming would be replaced by the practical computed delay time to achieve higher delay accuracy in the reflection USCT. The results from the numerical, phantom, and in vivo experiments show that our approach enables multi-modality imaging, accurate target localization, and precise boundary detection with the full-view fast imaging performance. The proposed method and its implementation are of great value for accurate, fast, and multi-modality USCT imaging, particularly suitable for highly acoustic heterogeneous medium.

Optimization of reconstruction time of ultrasound computed tomography with a piecewise homogeneous region-based refract-ray model

In this article, a novel ultrasound computed tomography (USCT) reconstruction algorithm for breast imaging is proposed. This algorithm is based on an ultrasound propagation model, the refract-ray model (RRM). In this model, the field of imaging is assumed as piecewise homogenous and is divided into several regions. The ultrasound propagation paths are considered polylines that only refract at the borders of the regions. The edge information is provided by B-mode imaging. Both simulations and experiments are implemented to validate the proposed algorithm. Compared with the traditional bent-ray model (BRM), the time of reconstructions using RRM decreases by over 90 %. In simulations, the imaging qualities for RRM and BRM are comparable, in terms of the root mean square error, the Tenengrad value, and the deformation of digital phantom. In the experiments, a cylindrical agar phantom is imaged using a customized imaging system. When imaging using RRM, the estimate of the phantom radius is about 0.1 mm in error, while it is about 0.3 mm in error using BRM. Moreover, the Tenengrad value of the result using RRM is much higher than that using BRM (9.76 compared to 0.79). The results show that the proposed algorithm can better delineate the phantom within a water bath. In future work, further experimental work is required to validate the method for improving imaging quality under breast-mimicking imaging conditions.

Radiopaque Iodosilane-Coated Lipid Hybrid Nanoparticle Contrast Agent for Dual-Modality Ultrasound and X-ray Bioimaging

Here, we report the synthesis of robust hybrid iodinated silica-lipid nanoemulsions (HSLNEs) for use as a contrast agent for ultrasound and X-ray applications. We engineered iodinated silica nanoparticles (SNPs), lipid nanoemulsions, and a series of HSLNEs by a low-energy spontaneous nanoemulsification process. The formation of a silica shell requires sonication to hydrolyze and polymerize/condensate the iodomethyltrimethoxysilane at the oil/water interface of the nanoemulsion droplets. The resulting nanoemulsions (NEs) exhibited a homogeneous spherical morphology under transmission electron microscopy. The particles had diameters ranging from 20 to 120 nm with both negative and positive surface charges in the absence and presence of cetyltrimethylammonium bromide (CTAB), respectively. Unlike CTAB-coated nanoformulations, the CTAB-free NEs showed excellent biocompatibility in murine RAW macrophages and human U87-MG cell lines in vitro. The maximum tolerated dose assessment was evaluated to verify their safety profiles in vivo. In vitro X-ray and ultrasound imaging and in vivo computed tomography were used to monitor both iodinated SNPs and HSLNEs, validating their significant contrast-enhancing properties and suggesting their potential as dual-modality clinical agents in the future.

Deep learning methods hold promise for light fluence compensation in three-dimensional optoacoustic imaging

SIGNIFICANCE

Quantitative optoacoustic imaging (QOAI) continues to be a challenge due to the influence of nonlinear optical fluence distribution, which distorts the optoacoustic image representation. Nonlinear optical fluence correction in OA imaging is highly ill-posed, leading to the inaccurate recovery of optical absorption maps. This work aims to recover the optical absorption maps using deep learning (DL) approach by correcting for the fluence effect.

AIM

Different DL models were compared and investigated to enable optical absorption coefficient recovery at a particular wavelength in a nonhomogeneous foreground and background medium.

APPROACH

Data-driven models were trained with two-dimensional (2D) Blood vessel and three-dimensional (3D) numerical breast phantom with highly heterogeneous/realistic structures to correct for the nonlinear optical fluence distribution. The trained DL models such as U-Net, Fully Dense (FD) U-Net, Y-Net, FD Y-Net, Deep residual U-Net (Deep ResU-Net), and generative adversarial network (GAN) were tested to evaluate the performance of optical absorption coefficient recovery (or fluence compensation) with in-silico and in-vivo datasets.

RESULTS

The results indicated that FD U-Net-based deconvolution improves by about 10% over reconstructed optoacoustic images in terms of peak-signal-to-noise ratio. Further, it was observed that DL models can indeed highlight deep-seated structures with higher contrast due to fluence compensation. Importantly, the DL models were found to be about 17 times faster than solving diffusion equation for fluence correction.

CONCLUSIONS

The DL methods were able to compensate for nonlinear optical fluence distribution more effectively and improve the optoacoustic image quality.

Computing a projection operator onto the null space of a linear imaging operator: tutorial

Many imaging systems can be approximately described by a linear operator that maps an object property to a collection of discrete measurements. However, even in the absence of measurement noise, such operators are generally "blind" to certain components of the object, and hence information is lost in the imaging process. Mathematically, this is explained by the fact that the imaging operator can possess a null space. All objects in the null space, by definition, are mapped to a collection of identically zero measurements and are hence invisible to the imaging system. As such, characterizing the null space of an imaging operator is of fundamental importance when comparing and/or designing imaging systems. A characterization of the null space can also facilitate the design of regularization strategies for image reconstruction methods. Characterizing the null space via an associated projection operator is, in general, a computationally demanding task. In this tutorial, computational procedures for establishing projection operators that map an object to the null space of a discrete-to-discrete imaging operator are surveyed. A new machine-learning-based approach that employs a linear autoencoder is also presented. The procedures are demonstrated by use of biomedical imaging examples, and their computational complexities and memory requirements are compared.