Computed ultrasound tomography in echo mode (CUTE) uses handheld pulse-echo ultrasound to image the spatial distribution of speed-of-sound. For a specific application of CUTE, the quantification of the fat fraction in the liver, convex probes are preferred to capture a large area of the liver in a sector scan. We demonstrate the adaptation of CUTE for such convex probes, showing phantom as well as in vivo results. We also discuss ideas of how to make use of the specific geometry of a convex probe in order to make such an implementation computationally efficient, and highlight the specific challenges when using convex compared to linear probes.

An estimate of more than 400,000 gastrostomy tubes (G-tubes) is placed annually in the United States. However, the poor organ visualization in endoscopy, ionizing radiation in fluoroscopy, US imaging artifacts, and limited point-of-care solutions limit the efficacy of image-guided G-tube placement procedures. Herein, we propose to develop a safe, point-of-care photoacoustic (PA) image-guidance system which utilizes a dual-wavelength approach for detecting the tissue’s endogenous and exogenous contrast agent for accurate G-tube placements. Our preliminary studies indicated that PA imaging accurately detected the dye-labeled colon-mimic, providing high contrast, artifact-free images of the introducer fiber superimposed on the US highlighted tissue background.

Ultrasound computed tomography techniques like full-waveform inversion are mathematically and computationally challenging, while existing codes are closed source, difficult to maintain, and slow to adapt to new research. Consequently, we present Stride, an open-source Python library for the solution of large-scale ultrasound tomography problems. Stride combines high-level, intuitive interfaces with high-performance wave-equation solvers based on Devito, a framework for automatic finite-difference code generation. We show how Stride can achieve state-of-the-art modelling accuracy and can switch seamlessly between 2D and 3D, while easily scaling from a workstation to a high-performance cluster.

Full wave 3D ultrasound tomography (3D-UT) is shown to be exactly congruent and ‘dual’ to training a gauge equivariant convolutional deep neural network (cDNN) with Lie group SO(2)xR applied to the weights (not feature vectors) and unusual ‘activation functions’. This explains high efficiency on NIVIDIA GPUs and indicates cross application of techniques from 3D-UT to cDNN and vice versa. ‘Backpropagation’ in both arenas is mathematically similar and the weights of the cDNN are the resulting image in 3D-UT. Data acquisition (DA) scenarios are interpreted as mappings from the Lie group manifold, and the 2-sphere, to the N-torus, with ramifications.

In this work, we proposed a novel normalized filed autocorrelation function (g1) –based functional ultrasound (fUS). We introduced a metrics derived from g1 function of ultrasound signals that is inversely proportional to blood flow and validated g1-fUS with in vivo experiments of whisker stimulation. Results show that g1-fUS is able to detect blood flow speed change in response to enhanced neural activity and is not affected by clutters compared to power Doppler based fUS, indicating that g1-fUS is a good candidate for whole brain functional imaging in small animals.

3D ultrasound tomography (3D-UT, or volography) gives heretofore unavailable resolution and accuracy in the 3D measurement of fibroglandular tissue in breast. The correct segmentation of the fibroglandular tissue requires an accurate high resolution speed of sound map which 3D UT (volography) produces and a robust segmentation algorithm. Our threshold based and Fuzzy C-means based algorithms have been compared and validated visually and with phantoms, and we show the effect of a row/column of dropouts on our reconstructions is minimal and indicate a variation of the volume of fibroglandular tissue in a variety of physiological situations for a select set of cases out of a larger 48 case cohort. We isolate glandular from ductal and structural tissue.

Prostate ultrasound imaging has utilized B-mode, however recent success in 3D ultrasound tomography (3D-UT) in the presence of bone, indicate using it to augment other potentially harmful or expensive modalities in clinic. Several fresh whole prostates were excised/inserted into bespoke polyacrylamide gel phantoms within 30 minutes of prostatectomy and scanned in the QT imaging scanner. The speed of sound (SOS) map resulting from the 3D-UT was used to create the refraction corrected reflection image compounded over 360 degrees resulting in sub-mm resolution. Several lesions were correlated with rigid transformations via anatomic landmarks with clinical MRI and H&E stained whole sections by experts in MRI and whole sectioning. Lesions were pointed out all 3 modalities and compared for multiple lesions indicating proof of concept of unique visibility of prostate lesions in 3D-UT (also volography) ex-vivo.