System and method for real-time global speed estimation with a single unfocused

Abstract

The inventors' framework constructs a loss metric using the complex autocorrelation of pre- beamformed RF data, and then fits this loss across a range of SoS candidates to find an optimal SoS value. We validated the inventors' framework's accuracy in vitro using agar staircase phantoms (SoS range: 1508-1682 m / s) and in vivo using svelte human calves (SoS range: 1573-1589 m / s). All validation data sets were paired with reference through-transmission measurements. On the in vivo validation data set, the inventors' framework outperforms two alternative single-transmit SoS estimation algorithms in both robustness (47-63% reduced failure rate) and estimation accuracy (2.3-24.7 m / s improved mean signed difference). The real-time feasibility of the inventors' framework was substantiated through a live demonstration at 20 fps on a portable ultrasound scanner. Overall, the inventors hope to facilitate the integration of SoS in tissue characterization, B-mode imaging, and other modes that are influenced by SoS such as Doppler estimation.

Description

SYSTEM AND METHOD FOR REAL-TIME GLOBAL SPEED ESTIMATION WITH A SINGLE UNFOCUSEDTRANSMISSIONCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] The present application claims the benefit of U.S. Provisional Application No. 63 / 697,475 filed on September 21, 2024, entitled SYSTEM AND METHOD FOR REALTIME GLOBAL SOUND SPEED ESTIMATION WITH A SINGLE UNFOCUSED TRANSMISSION, and the benefit of U.S. Provisional Application No. 63 / 783,556 filed on April 4, 2025, entitled SYSTEM AND METHOD FOR REAL-TIME GLOBAL SOUND SPEED ESTIMATION WITH A SINGLE UNFOCUSED TRANSMISSION, which are hereby incorporated by reference in their entirety.FIELD

[0002] This disclosure relates to a system and method for tissue speed-of-sound (SoS) estimation.BACKGROUND OF THE INVENTION

[0003] In biomedical ultrasound imaging, the tissue speed-of-sound (SoS) is essential for localization of received radiofrequency (RF) echoes after a pulsing event. Knowledge of tissue SoS also has potential clinical impact as a biomarker for tissue pathologies. In recent years, pulse-echo methods for estimating tissue SoS have been developed. However, these methods mostly rely on leveraging interactions from multiple steered transmissions to calculate the SoS.

[0004] What is needed is an improved system and method which can address some of the limitations in the prior art.SUMMARY OF THE INVENTION

[0005] In this disclosure, the inventors propose what is believed to be the first experimentally validated, real-time "one shot, one SoS" solution that leverages the pre-beamformed channel data acquired from a single unfocused transmission to derive a global SoS estimate. This new single-shot global SoS estimator is based on the design of a loss minimization strategy that aims to compare, over a range of candidate SoS values, the sum of normalized autocorrelation coefficients (SNAC) obtained from the pre-beamformed channel data at different pixels in the imaged view. As will be shown, this new framework has demonstrated accurate SoS estimation performancein vitro and in vivo, as well as real-time implementation feasibility in parallel to B- mode imaging on a portable ultrasound research scanner.

[0006] In an embodiment, a novel global SoS estimation algorithm is proposed that requires only a single steered plane wave transmission. The proposed framework constructs a loss metric using the complex autocorrelation of pre-beamformed RF data, and then fits this loss across a range of SoS candidates to find an optimal SoS value. The framework's accuracy was validated in vitro using agar staircase phantoms (SoS range: 1508-1682 m / s) and in vivo using svelte human calves (SoS range: 1573- 1589 m / s). All validation data sets were paired with reference through-transmission measurements. On the in vivo validation data set, the proposed framework outperforms two alternative single-transmit SoS estimation algorithms in both robustness (47-63% reduced failure rate) and estimation accuracy (2.3-24.7 m / s improved mean signed difference). The real-time feasibility of the proposed framework was substantiated through a live demonstration at 20 fps on a portable ultrasound scanner. Overall, the proposed framework can facilitate the integration of SoS in tissue characterization, B-mode imaging, and other modes that are influenced by SoS such as Doppler estimation.BRIEF DESCRIPTION OF DRAWINGS

[0007] FIG.l shows an overview of an SoS estimation framework in accordance with an illustrative embodiment.

[0008] FIG.2 shows RF data from a single plane wave transmit alongside corresponding |3-mode images.

[0009] FIG.3 (a)-(c) summarize the results of applying the SoS estimation framework ofFIG. 1 on an in vitro agar.

[0010] FIG.4 (a)-(f) summarize the results of applying the SoS estimation framework of FIG. 1 on the more challenging in vivo human calf scenarios (Sec. 3.2).

[0011] FIG.5 (a)-(b) show snapshots of the SoS estimation framework in live operation both in an in vitro (FIG. 5(a)) and an in vivo (FIG.5 (b)) scenario.

[0012] FIG.6 shows a generic computing environment which may be suitable for providing a platform for one or more embodiments.

[0013] FIG. 7 shows an SoS estimation performance comparison on in vitro agar phantoms while using focused transmissions and the present proposed method (a), speckle brightness maximization (SBM) (b), and radiofrequency coherence (RFC) (c).DETAILED DESCRIPTION OF THE INVENTION

[0014] Ultrasound, as a pulse-echo imaging modality, relies on the emission of acoustic pulses and the subsequent reception of the reflected echoes [Perrot 2021], Such a pulse-echo process is fundamentally influenced by the speed-of-sound (SoS), an acoustic property that describes the speed at which the pulse propagates through a medium [Perrot 2021], The tissue SoS is not only essential for proper image formation [Anderson 2000], but also has potential as a biomarker for tissue pathologies such as tumour malignancy or liver steatosis [Li 2009, Telichko 2022, Stahli 2023], Conventionally, the tissue SoS was measured via pitch-catch or reflection-type devices that had limited applicability for in vivo human assessment [Bamber 1979, Sehgal 1986], However, recent pulse-echo SoS measurement frameworks have incorporated simultaneous imaging capabilities to enhance the clinical potential of SoS as a biomarker for tissue health diagnostics [Jaeger 2015, Xiao 2024],

[0015] Pulse-echo SoS measurement has grown in popularity with several proposed methods for global, or average, SoS estimation as well as local SoS mapping [Krucker 2004, Jaeger 2015, Jakovljevic 2018, Sanabria 2018, Xiao 2024], These methods, which are implemented using clinical ultrasound transducers, have demonstrated the practicality of measuring the tissue SoS in real-time at the bedside [Heriard-Dubreuil 2023, Stahli 2023, Xiao 2024], Global SoS methods provide an average SoS of the medium to facilitate high-level tissue characterization of the imaged medium [Krucker 2004, Xiao 2024], While these methods can be simple to implement, the inherent lack of localization limits their broader application for more inhomogeneous imaging scenarios [Xiao 2024], Local SoS methods can provide a SoS image of the imaged medium to better distinguish between tissue types for inhomogeneous media [Jaeger 2015, Sanabria 2018], These methods have been used to tackle pathological scenarios, such as assessing liver pathologies [Telichko 2022, Stahli 2023], However, these algorithms demand increased computation whilst also requiring many transmissions for robust SoS mapping [Sanabria 2018, Stahli 2020, Ali 2022], Regardless of global or local SoS estimation, the majority of recent pulse-echo SoSestimation algorithms rely on pulsing events from multiple transmission paths to derive the SoS estimate [Jaeger 2015, Sanabria 2018, Ali 2022, Ali 2023, Bezek 2023, Stahli 2023, Xiao 2024], This multi-pulse transmission approach remains a potential barrier to increased adoption of current SoS estimation frameworks as changes to the imaging methodology and transmit sequence may be required for integration of SoS information.

[0016] In this disclosure, the inventors propose a novel global SoS estimation algorithm that requires only a single unfocused transmission. The inventors' framework functions by comparing the channel-wise consistency of the underlying reflected signal using the pre-beamformed RF data for several assumed SoS values. This framework was validated experimentally both in vitro and in vivo with reference through-transmission SoS measurements. The inventors additionally compared the inventors' proposed algorithm with two other potential approaches for global SoS estimation requiring only a single transmission event. Lastly, the inventors demonstrated the real-time capabilities of the inventors' framework with a live demonstration on a portable ultrasound scanner. As the inventors' method uses a single plane wave transmission, the proposed method can expand the range of ultrasound applications where SoS information can be extracted and used.

[0017] Now referring to FIG. 1, an overview of the inventors' SoS estimation framework is shown. Data is first acquired from a single transmit (step 1). Next, the RF samples from each channel are appropriately delayed and selected without summation (step 2) for all pixels and all SoS candidates. A loss metric based on the complex integral of the normalized autocorrelation is evaluated (step 3) and then a fitted loss is used to select the minimum to declare as the SoS estimate (step 4).

[0018] The framework structure is enhanced from a previous work on real-time SoS estimation from the inventors' group, with the fundamental distinction of requiring only a single unfocused transmit instead of multiple steered transmits for real-time SoS estimation [Xiao 2024], First, the inventors acquire the raw radiofrequency (RF) data from a single plane wave transmission and preprocess the raw RF data via bandpass filtering around the transmit frequency and analytic signal conversion via the Hilbert transform (Fig. 1, step 1). Next, for a selection of SoS candidates, the inventors apply only the geometrical delay step of the classical DAS algorithm andaccount for the refraction-like effect at the probe-media interface [Xiao 2024] for every pixel position to. In doing so, the inventors can extract each pixel's time-delayed channel data ensemble, that contains the corresponding data sample of every array channel after time delay compensation; coherent summation of this ensemble is typically done to form a beamformed pixel value of the ultrasound image (Fig. 1, step 2) [Perrot 2021], This pre-beamforming step is applied to the 2D RF data (samples by elements) results in a 3D tensor structure (axial pixels by lateral pixels by elements) that contains the relevant RF data for that pixel. Black regions (FIG. 1, step 2) indicate that data from certain elements is not used for beamforming some pixels due to F- number considerations [Perrot 2021],

[0019] From the acquired RF vector (elements by 1) at each pixel location, a loss function based on the complex autocorrelation is calculated and then integrated over the domain. This loss function is, in principle, minimized when the beamforming SoS matches the medium SoS. Next, these losses are summed across all pixels and used to construct a SoS-dependent metric (FIG. 1, step 3). A smoothing spline is then fit to this loss metric based on the pre-selected SoS candidates and the SoS estimate produced by the inventors' framework is the minimum of the fitted loss (FIG. 1, step 4). The subsequent subsections delve into the principles behind the inventors' overall approach.Considerations for Loss Function Design

[0020] In devising a robust single-shot SoS estimator, the loss function needs to be carefully defined to distinguish the matched inter-channel signal features attributed to an accurate SoS from mismatched ones stemming from an inaccurate SoS. This principle is pedagogically illustrated in Fig. 2, which shows the effect of inaccurate SoS during geometrical delay calculations in a case of a wire target placed in water whose SoS is 1489 m / s. Here, the channel RF data matrix from a single -15° plane wave transmit (Fig. 2(a)) is shown alongside the corresponding B-mode images (Fig. 2(b)-(d); 3x3 mm2field-of-view around the wire) for three candidate SoS values (1389 m / s, 1489 m / s, 1589 m / s). Also shown in this figure are the appropriately delayed (not yet summed) RF data envelope in a neighborhood of the wire target (FIG. 2(e)-(f); geometric receive delays calculated based on position of orange pixel). Thedefocusing of the wire target in the B-mode images (FIG. 2(b)-(d)) is an anticipated effect of SoS mismatch that has been explained in previous works for both focused and unfocused imaging [Anderson 2000, Xiao 2024], Similarly, the RF data alignment of the ideal hyperbolic patterns (FIG. 2(e)-(f) has been examined when varying the SoS used for beamforming [Ali 2023],

[0021] We have replicated these effects in FIG. 2 (a)-(g) to illustrate how either the post-beamformed or pre-beamformed data can be exploited for SoS estimation. Here, a) RF data from single plane wave transmit corresponding to wire target in water, (b)-

[0022] (d) beamformed images corresponding to varying SoS values used for delay-and- sum. (e)-(f) appropriated delayed RF data around the vertical line passing through the highlighted point (orange dot) for each SoS value.

[0023] Fundamentally, SoS mismatch between an assumed beamforming SoS and the actual SoS of the underlying medium leads to pronounced differences in the similarity of the pre-beamformed data samples differences in the similarity of the pre- beamformed data samples between array channels at a given pixel position. A loss function for SoS estimation may then be defined based on an assessment of such similarity, as will be described below. In the inventors' work, the inventors propose a novel metric that captures the inherent defocusing attributed to inaccurate SoS and apply it as the loss function of the inventors' single-shot SoS estimator. Specifically, to capture SoS-relevant information, the inventors' new metric is defined as the total SNAC of all time-delayed channel data ensembles within the region of interest. For a given pixel p=(z,x), where z and x are the axial and lateral pixel positions respectively, let the geometrically delayed RF vector (sized N_Exl, the number of elements / channels on the probe) at SoS c for that pixel p be defined as (r_p ) (c). Then, the loss function, which is essentially the negated total SNAC for all pixels p G P of each beamforming SoS candidate c is mathematically expressed as: (1>where p(v ”',l) is the complex autocorrelation function of a vector v at lag I, and the inner summation is essentially the SNAC of a particular time-delayed channel data ensemble.Note that p(y, Z) is mathematically defined as follows for a predefined set of lags L = {Z G Z, Zmin< I < Zmax}.where v(Z) denotes the ithelement of vector v, and * denotes the complex conjugation. Demonstration code for applying the negated total SNAC as the loss function for SoS estimation is provided online (https: / / github.com / litmus- uwaterloo / Single-Shot-Sound-Speed).

[0024] Intuitively, when the beamforming SoS (used to calculate the geometrical delays) does not match the actual medium SoS, the pre-beamformed samples across the receiving elements would differ in both amplitude and phase from each other. Conversely, when the beamforming SoS matches, the correlation between the complex-domain samples (having both amplitude and phase) will be maximized. In theory, this relationship still roughly holds when considering signal changes due to attenuation, directivity signal loss, and noise.

[0025] To increase the robustness of the loss function, the inventors can restrict its application to only a subset of the pixels p G P that have relatively high post- beamformed intensity when beamformed using a candidate SoS. Specifically, for each candidate SoS, after deriving the time-delayed channel data ensembles for each pixel 7j / (c, a) and computing its SNAC, the pixels can be ranked based on the mean element-wise intensity of their ensembles. The total SNAC would then only be summed from the SNACs of the top-ranked pixels (e.g., the top 80%) to suppress spurious contributions from low-intensity pixels.

[0026] Our single-shot SoS framework can naturally be extended to multiple steered transmissions if they are available. Through the additional multi-angle information, the robustness of the system can be enhanced. The inventors propose two methods for extending the inventors' framework in multi-angle scenarios. First, the inventors can estimate the SoS for each angle a in the transmission scheme A separatelythrough Eq. (1) and then pool / r{-} the estimated SoS's into a single estimated SoS.The inventors denote this extension as MTE-IA (multi-transmit extension with independent angles):Second, the inventors can form a multi-transmit 3D data matrix by concatenating (— ~) the sampled 3D data matrix for each transmission across the channel dimension and then applying the proposed algorithm on the resulting tensor as if it was constructed from a single transmission. The inventors denote this second extension as MTE-C (multi-transmit extension with concatenation):

[0028] The inventors' framework was validated using both in vitro and in vivo data collected from open-platform research ultrasound scanners alongside a custom pitchcatch SoS measurement device. For the in vitro data, the inventors re-used data collected for a previous SoS study [Xiao 2024] using the SonixTouch (Analogic Ultrasound; Peabody, MA, USA) device equipped with an L14-5 linear array (Analogic Ultrasound; 128 Tx / Rx elements; 5 MHz transmit; -15O:1°:15° steering; 1 pulse cycle). For in vivo data, data was acquired using a US4R-Lite portable research scanner (US4US; Warsaw, Poland) using the same L14-5 array and parameters. This in vivo data was secondary use of musculoskeletal ultrasound data originally collected from the calves of human volunteers for a previous study [Xiao 2024], Only data from participants who originally consented to secondary use of their data were included as part of this work (University of Waterloo Research Ethics Board Protocol No. 44778). Reference SoS values for all data were measured using a previously described custom pitch-catch system with validated accuracy [Xiao 2024],Data Description for Validation of Framework

[0029] To characterize the SoS estimation accuracy of the inventors' proposed framework, the SoS estimates output by the inventors' framework were compared to the reference SoS measurements from the custom pitch-catch device. Validation in vitro was performed on agar staircase phantoms that were constructed in-house using varying glycerol concentration (2-35.6% by weight) to control the phantom SoS. In total fifteen phantoms were used for comparison (reference SoS range: 1508-1682 m / s). Pulse-echo data was collected from the largest step of the staircase, while reference SoS was gathered across 15 total measurements from the three steps of each staircase. Validation in vivo was performed via human calf data (both left and right appendages). The custom pitch-catch device was placed across the gastrocnemius muscles of each volunteer and the average of 5 measurements was used as the reference SoS value (reference SoS range: 1573-1589 m / s). The inventors restricted the data that was used for validation of the inventors' framework based on the thickness of the subcutaneous adipose tissue of the participants. To mitigate the known effects of inhomogeneity on global SoS estimator, only imaging views with less than 8 mm of subcutaneous adipose tissue were used for the assessment of the inventors' framework's accuracy [Xiao 2024], In total, 930 calf acquisitions from a set of 23 volunteers, consisting of medial and lateral views from both the left and right legs, were re-used in this work to evaluate the SoS estimation performance.Framework Comparison with Alternative Speed-of-Sound Estimation Frameworks

[0030] Our proposed framework was additionally compared against two SoS estimation frameworks that also have the capability to operate on single transmit data. The first method was a speckle brightness maximization (denoted SBM) approach for a homogeneous region of speckle [Anderson 2000], The framework the inventors used was adopted from the more recent re-implementation of the algorithm where open access code was provided as part of an ultrasound deep learning challenge [Hyun 2021], The inventors adapted that implementation to function on single-angle beamformed images instead of coherently compounded images. The metric used as the loss function of this image-domain SoS estimator can be expressed in a pixel-wise fashion (across pixels p) as:As shown in Eq. (5), instead of using pre-beamformed channel datafor SoS estimation (as is used in the inventors' new method; see Eq. (1)), the SBM approach instead uses the post-beamformed pixel intensity Ip(beamformed at SoS candidate c for transmit angle a)

[0031] The second method was an RF coherence (denoted RFC) approach that minimizes the phase dispersion [Perrot 2021], This method was inspired by a similar, earlier method that examines the phase variance of RF channel data [Yoon 2011], For comparison against the inventors' proposed method, the inventors again adapted code for SoS estimation (ezsos function) that was publicly provided [Perrot 2021] while making the appropriate parameter adjustments for the faithful comparison of the method, as it's been demonstrated to work on multi-transmit data.. Explicitly stating this loss function in similar format to Eq. (1) and Eq. (5), the inventors arrive at the expression:Where o(-) is the complex standard deviation (a CA(N_C )i->R function), cf(’) extracts the unwrapped phase of the complex vector (r_p ) (c,a), and p(-) is the complex arithmetic mean. This method is similar to the inventors' proposed framework in its use of pre-beamformed channel RF data but leverages an alternative effect of the SoS mismatch in the RF data.

[0032] Our single-shot SoS estimator and the two alternate methods were evaluated on the same sets of in vitro and in vivo data (Sec. 3-2). For all methods, the -12° and - 3° angles were selected for comparison. For all results, the same pixel grid (128x128, axial: 12-36 mm, lateral: -11-9 mm) and F-number (1.4) were used. For the inventors' proposed framework, l_min and l_max were set to be -N] _E / 2 and N_E / 2 respectively.

[0033] Additionally, the effect of utilizing a subset of sampled pixels for SoS estimation was explored for all methods, as some of the methods explicitly require to be applied on "high intensity regions only" [Hyun 2021] and the inventors' proposed total SNAC metric and RFC impose an upper bound on the loss of a given pixel. For each explored percentage of pixels used for SoS estimation, the metric was only applied to the top assessed pixels (ranked by beamformed intensity). To minimize the effects of implementation variance, the grid was stretched axially by 0,2,4 and 6mm and the best performing grid was used for the reported accuracy and consistency. A subset of the explored percentage range that best describes the peak performance operating scenarios for all methods is then reported.

[0034] The two multi-angle extensions were evaluated using the same in vitro dataset as previously described. Using MTE-AIA, the inventors evaluated both the accuracy for each angle for the framework as well as the accuracy of pooling the SoS's from the steered transmissions. The pooled SoS was a trimmed mean where the inventors cut out the 20% (2 of the 10 estimated SoS) of the outliers to the mean SoS.

[0035] Using MTE-C, the inventors evaluated the accuracy and precision of the single SoS estimate that arose from the modified 3D data matrix (Sec. 2.4). In practice, since the F-number was set at 1.4 (Sec. 3.3), the length of (r_p ) (c,a) is not consistent for all sampled pixels, so instead an oversized 3D data matrix was pre-allocated and filled with the corresponding samples to ensure no gaps in-between the concatenated ensembles (r_p ) (c,a_i ), for this implementation of MTE-C, the channel data ensembles were concatenated in the ascending angle order, i.e. a_(i+l)>a_i, iG N,i<k.Real-time System Implementation

[0036] Our proposed framework was implemented on a portable ultrasound scanner (US4R-Lite) to demonstrate the real-time feasibility of the algorithm. Our hardware setup consisted of the same US4R-Lite scanner that was connected via Thunderbolt USB-C to a laptop (ThinkPad XI Extreme Gen 4; Lenovo; Beijing, China) with discrete GPU (RTX 3080 Laptop; Nvidia; Santa Clara, CA, USA). Programming was performed in Python (ver. 3.8.5) and GPU acceleration was applied via the CuPy library (ver. 10.6.0) with CUDA (ver. 11.2) backend in conjunction with the US4R Python API (US4US; Warsaw, Poland).

[0037] For efficient selection of RF samples, the inventors used a sparse matrix beamforming (SMB) approach [Xiao 2024, Hou 2014, Kamimura 2021], Specifically, the inventors encoded the selection of RF data for all SoS candidates into a single matrix-multiply operation that was supported by the CuPy GPU library's sparse matrix functionality. Prebuilt CuPy functions were also used to accelerate the calculation of the autocorrelation via the fast Fourier transform and to accelerate the selection of high-magnitude post-beamformed pixels via parallelized sorting. The inventors' realtime implementation used the top 90% of pixels (in terms of magnitude) for loss function calculation and used the same 1.4 F-number aperture as in the previous subsection. To produce the SoS estimation, the inventors fitted a smoothing spline and declared the minimum of the smoothed loss as the global SoS estimate. Finally, to stabilize the displayed SoS to the user, the inventors displayed the rolling average within a window of 10 frames; this rolling average was calculated after trimming the highest and lowest values to reduce the effect of outliers.

[0038] Using the real-time prototype of the inventors' framework, the inventors measured the framerate timings and assessed the robustness during live performance. A cineloop was recorded during live operation with the inventors' framework. The inventors imaged a standardized multi-purpose phantom (Multi- Purpose Multi-Tissue Ultrasound Phantom; CIRS Inc., Norfolk, \ZA, USA) with a nominal SoS of 1540 m / s, two newly constructed agar-glycerol phantoms with measured SoS of 1595 m / s and 1539 m / s, and one human calf with measured through-transmit SoS of 1588 m / s.Results

[0039] FIG. 3(a)-(c) summarize the results of applying the inventors' framework on the in vitro agar staircases. SoS estimation performance comparison on in vitro agar staircase phantoms between the inventors' proposed method (a), speckle brightness maximization (SBM) (b), and radiofrequency coherence (RFC) (c).

[0040] Our proposed method demonstrated high accuracy compared to the alternative methods, which is quantified by the mean signed difference (MSD) between the estimated and the reference SoS values, with MSD of 0.4m / s (o: 6.5m / s) for the -12° angle and 0.5m / s (o: 9.6m / s) for the 3° angle. In contrast, the SBM and RFC methods,using the same data, achieved MSD of 3.2m / s (o: 6.9m / s.) and 10.5m / s (o: 12.8m / s) (FIG. 3(b)) and 2.6 / s (o: 47.5m / s.) and 0.6m / s (o: 47.7m / s) (FIG.3 (c)). Cases where any algorithm resulted in a SoS estimate that was equal to the minimum or maximum SoS candidate were declared failure cases and were plotted at the top edge of the graph, as no clear minimum loss was established (orange downward-pointing triangles). Only non-failed estimates were used for calculating the MSD. The inventors' proposed method resulted in zero failure cases on all agar phantoms for both transmit angles, while SBM and RFC only produced a non-failed SoS value in 80% and 90% of the cases, respectively.Accurate and Robust Estimation for In Vivo Human Calves

[0041] FIG. 4(a)-(f) summarize the results of applying the inventors' framework on the more challenging in vivo human calf scenarios (Sec. 3.2). The inventors' proposed method demonstrated high accuracy compared to the alternative methods, with a quantified MSD of 3.8m / s (o: 15.7m / s) and 2.5m / s (o: 22.5m / s) for the 12° and 3° angles, respectively. In contrast, the SBM and RFC methods achieved MSD of 6.1m / s (o: 34.0m / s.) and 22.0m / s (o: 32.1m / s) (FIG. 4(b)(e)) and 15.4 / s (o: 25.6m / s.) and 29.5m / s (o: 34.2m / s) (FIG.4 (c)(f )), respectively. Only non-failed estimates were used for calculating the MSD. Assessing the non-failure cases, the inventors' method resulted in a genuine SoS estimation on the human data acquisitions in 74% and 74% of the 12? transmit and 3? transmit cases, respectively. The successful estimation rates of the alternative SBM method were only 27% and 18% and of the RFC method only 10% and 50%, respectively for the two compared angles.

[0042] Table 1 (red) summarizes the effect of utilizing different inclusion percentages of ranked pixels (with pixels being ranked differently for each SoS candidate) within the beamformed grid on the estimation accuracy, characterized by the achieved MSD of each of the evaluated algorithms. Accordingly, Table 1 (green) summarizes the percentage of convergence (PC) instances where each framework was able to converge (PC) to a SoS value that was strictly within the tested range upper and lower bounds. Overall, the inventors' method was able to better sustain the high accuracy performance and reliability over a wider range of used pixels, when compared to the alternative methods.Table 1. Summary of the performance of the tested frameworks under decreasing number of oixels used for SoS estimation at 12° Tx. Hiahliahted for each method is the oixel oercentaae

[0043] A snapshot of the inventors' framework in live operation is highlighted in FIG. 5 for both an in vitro (FIG. 5(a)) and an in vivo (FIG.5 (b)) scenario. These snapshots correspond to Movie 1, which is provided as part of the supplementary material. The live framerate of the system was 20 fps. Qualitatively, the trimmed averaging of the SoS estimates increased the robustness of the displayed SoS as failure cases did not contribute significantly to the displayed SoS.

[0044] Now referring to FIG. 6, shown is a schematic block diagram of a generic computing device that may provide a suitable operating environment in one or more embodiments. A suitably configured computer device, and associated communications networks, devices, software, and firmware may provide a platform for enabling one or more embodiments as described above. By way of example, FIG. 6 shows a generic computer device 600 that may include a central processing unit ("CPU") 602 connected to a storage unit 604 and to a random-access memory 606. The CPU 602 may process an operating system 601, application program 603, and data 623. The operating system 601, application program 603, and data 623 may be stored in storage unit 604 and loaded into memory 606, as may be required.Computer device 600 may further include a graphics processing unit (GPU) 622 which is operatively connected to CPU 602 and to memory 606 to offload intensive image processing calculations from CPU 602 and run these calculations in parallel with CPU 602. An operator 610 may interact with the computer device 600 using a video display 608 connected by a video interface 605, and various input / output devices such as a keyboard 610, pointer 612, and storage 614 connected by an I / O interface 609. Inknown manner, the pointer 612 may be configured to control movement of a cursor or pointer icon in the video display 608, and to operate various graphical user interface (GUI) controls appearing in the video display 608. The computer device 600 may form part of a network via a network interface 611, allowing the computer device 600 to communicate with other suitably configured data processing systems or circuits. A non-transitory medium 616 may be used to store executable code embodying one or more embodiments of the present method on the generic computing device 600Discussion and Conclusion

[0045] The tissue SoS is essential for ultrasound image formation with additional applications as a quantitative tissue biomarker. Typical pulse-echo methods for SoS estimation rely on leveraging information across multiple transmissions for robust calculation of tissue SoS., and implementing these algorithms on an ultrasound scanner would fundamentally require an ultrasound imaging sequence redesign that may not be easily achievable on clinical scanners with embedded system architectures [Powers 2011], In this work, the inventors described a real-time SoS estimation algorithm that required only a single unfocused transmission, and can be implemented with a system's existing imaging sequence. The inventors' innovation relies on the behavior of the complex autocorrelation across geometrically delayed RF data when there is SoS mismatch (Sec. 2.3). Through secondary use of data from a previous study [Xiao 2024], the inventors validated the SoS estimation accuracy of the inventors' single plane wave algorithm in comparison to reference through- transmissions SoS measurements with a custom pitch-catch device using agar phantoms (Fig. 3) and in vivo measurements of the human calf (Fig. 4). Lastly, the inventors also demonstrated that the inventors' technique can straightforwardly be readily implemented for real-time operation (Fig. 6, Movie 1).

[0046] In the inventors' experiments, all three compared methods exhibited higher SoS estimation accuracy at 12° steering angle as compared to 3° steering (Fig. 3 and 4). This difference in performance between steering angles is most likely due to the refraction-like effect at the probe-media interface, making it such that a simple delay compensation is insufficient to capture the time-of-flight discrepancy that occurs at aspecific depth. All tested SoS estimators also benefitted from the restriction of the loss metric to higher intensity pixels. Note that the inventors' proposed method and the RFC method are intensity normalized and can, unintentionally, overly weigh noisy pixels or regions without echogenic structure. Similarly, SBM relies on the coherence of strong scatterers to outweigh the energy averaging effect when using different candidate SoS for beamforming. Thus, considering only the highest intensity pixel led to increased performance of the SBM metric as it maximized the strength of the post beamformed pixels. The inventors noted that, in practice, the intensity cutoff of pixels to exclude for optimal performance was more generous for the inventors' proposed method compared to the alternatives.

[0047] When the proposed method was extended into the multi-transmission regime following MTE-AIA and MTE-C (Sec. 3.4) both accuracy and precision were naturally improved, as the information content used for SoS estimation increased. Additionally, MTE-ANevertheless, the performance of the inventors' single-shot implementation (Fig. 3a) was not substantially worse than that of MTE-IA and MTE-C (Fig. 5b and 5c). The inventors' results indicate that the new single-shot SoS estimator enjoys the unique advantage of readily integration with a conventional ultrasound imaging sequence (fundamentally not achievable with a multi-shot SoS estimator) without sacrificing major loss of estimation accuracy and real-time operability. Regarding the specific performance of the multi-transmit extensions, MTE-IA revealed the varying nature of the estimation error for each of the steered transmission angles (Fig. 5(a)). MTE-C (Fig. 5(c)) proved to be the superior framework out of the two proposed extensions. The inventors believe the improved robustness arose from leveraging the inter-dependencies within the multi-transmission set of RF data, as opposed to considering each of the transmission events as an independent observation (as in MTE-A, Fig. 5(b)). Furthermore, when compared to the inventors' previous work on the same dataset [Xiao 2024], which achieved an MSD of -1.8m / s (o: 2.2m / s), the MTE-C achieved higher accuracy and slightly improved standard deviation. One possible explanation for the difference in the inventors' framework leveraging pre- beamformed RF data as compared to post-beamformed analytic images; however, there is a potential trade-off in the required computation cost.lA, Fig. 5(b)).

[0048] Our work compared favorably to alternative SoS estimation frameworks that could use only a single transmission (FIG. 3, FIG. 4). Both the speckle brightness maximization and RF coherence approaches (Sec. 3.3) resulted in an increased number of failure cases for the algorithm, particularly for the more challenging in vivo imaging scenarios. Compared to the RF coherence (RFC) approach (Eq. (3)), the inventors' proposed loss function (Eq. (2)) across the pre-beamformed RF data was better suited for sample comparison across unevenly weighted receive hyperbolas across elements (due to factors such as attenuation or directivity). In the inventors' experiments, all three compared methods benefitted from the restriction of the loss metric to higher intensity pixels. The inventors' proposed method and the RFC method are intensity normalized and can, unintentionally, overly weigh noisy pixels or regions without echogenic structure. Similarly, SBM relies on the coherence of strong scatterers to outweigh the energy averaging effect when using different candidate SoS for beamforming. Thus, considering only the highest intensity pixel led to increased performance of the SBM metric as it maximized the strength of the post beamformed pixels. The inventors noted that, in practice, the intensity cutoff of pixels to exclude for optimal performance was more generous for the inventors' proposed method compared to the alternatives.

[0049] As the proposed framework operated robustly despite only requiring a single unfocused transmission, the framework is well-positioned for integration into other high-frame-rate or ultrafast ultrasound imaging paradigms [Tanter 2014], For example, multi-angle vector flow imaging, which uses one to five steered transmits [Yiu 2016], can potentially benefit from using single-angle transmissions for initial SoS estimation prior to beamforming as the SoS has a known deteriorative effect on the accuracy of Doppler velocity estimates [Christopher 1995], The proposed algorithm, as a global SoS estimator, would in principle suffer from similar drawbacks as other global estimators [Xiao 2024], However, the pixel-wise nature of the algorithm can, via previously demonstrated means [Jakovljevic 2018], be extended by varying the selected pixel region to achieve a more localized SoS estimation. Overall, the low requirements and simplicity of the inventors' proposed technique highlight the simplicity of integrating SoS estimation into high-frame-rate ultrasound frameworksfor imaging improvement, while simultaneously increasing the clinical potential of SoS as a tissue characterization biomarker for tissue health diagnostics.

[0050] In an embodiment, the proposed method is effective not only for unfocused transmissions, but also for focused transmissions as well. The performance of the present method can be evaluated when estimating the global Speed of Sound (SoS) while using a single focused transmission. A dataset can be devised consisting of 6 different agar-glycerol phantoms with measured reference SoS varying from 1516 m / s to 1652 m / s. A single-cycle focused transmission with focal point located at the lateral center of the L14-5 linear transducer and at an axial depth of 2 cm from the transducer face was used to estimate the SoS of each sample. A beamforming grid spanning from 12 mm to 45 mm axially and from -19 mm to 19 mm laterally was used to sample the points required for SoS estimation for each of the evaluated methods. The pixel positions were beamformed based on the transmission geometry of the focused transmit (resulting in an 'hourglass' shape of the insonified pixels).

[0051] Similar to the experiments for unfocused transmissions, a grid search approach for determining the percentage of pixels to be used for SoS estimation was followed to find the percentage of pixels that would yield the best accuracy for each of the 3 methods. 20% of beamformed pixels were used in the inventors' method (corresponding mostly to the insonified pixels), in contrast, only 0.3% were used for the other methods.

[0052] The present method demonstrates superior accuracy and precision (results summarized in FIG. 7), denoted by a lower mean arithmetic error (MAE) (MAE: -1.2 m / s, o: 6.3 m / s) when compared to Speckle Brightness Maximization (MAE: 24.3 m / s, o: 10.5 m / s) and Radio Frequency Coherence (MAE: 1.1 m / s, o: 15.0m / s). Overall, this demonstrates that the inventors' proposed framework is capable of accurately estimating the average SoS of a medium by using a single focused transmission.Illustrative Use Cases

[0053] Medical Ultrasound - As an illustrative use case, our technology can be applied in the field of medical ultrasound. The speed-of-sound has been shown to be an effective tissue biomarker, differentiating between tissue pathologies such as malignant or benign tumours and between steatotic livers and healthy livers. Morespecifically, global (average) speed-of-sound estimation has demonstrated the ability to quantify tissue composition as the global speed-of-sound is highly correlated with fatty infiltration in the muscles and breast density. Our technology operates via the pulse-echo paradigm using conventional ultrasound transducers, allowing for point- of-care operation. We envision that the technology can be used in clinical settings to quickly evaluate or monitor the tissue composition (e.g. if the breast demonstrates higher sound speed, then the individual can be flagged for follow-up checkup) (e.g. if the muscle demonstrates lower sound speed after long ICU stay, then the individual can be flagged for additional physiotherapy regime for recovery)

[0054] Non-destructive Testing - As another illustrative use case, the speed-of-sound is a fundamental acoustic property of materials. Knowledge of the speed-of-sound can be used to evaluate and monitor material composition. For example, the speed-of- sound changes as materials such as resin or concrete cure. Our technology can be used to identify when a material is fully cured. Alternatively, our technology can also potentially be used to evaluate the condition of materials such as concrete. E.g. if we identify the optimal condition of concrete to have a specific speed-of-sound, then deviations from that speed-of-sound can indicate possible structural issues. Our technology can also be applied to potentially identify unknown materials. Different metal alloys or liquid mixtures can also have different speed-of-sound values. Using an ultrasound array transducer on unknown metals or liquids can potentially help to reveal the composition of the material.

[0055] Meat and Livestock Quality - As yet another illustrative use case, the speed-of- sound is known to be related to intramuscular fat, which is essentially marbling in the meat context. Our device can be used to evaluate the intramuscular fat through the speed-of-sound (high speed-of-soundless marbling, low speed-of-soundhigh marbling). This technology can be used noninvasively on livestock, carcasses, and processed meats using ultrasound array transducers. For livestock, our technology can potentially guide livestock selection for breeding purposes, or readiness of livestock for slaughter. For carcasses, our technology can guide processes such as pricing, distribution, and automatic sorting based on the speed-of-sound as a proxy for the meat quality. For processed meats, our technology can be used to monitor products for quality control.

[0056] Thus, in an aspect, there is provided a computer-implemented method of estimating a speed-of-sound (SoS) of a tissue medium in real-time, comprising:(i) generating a single acoustic transmission, the transmission being unfocused or focused;(ii) acquiring pre-beamformed radiofrequency (RF) channel data responsive to the transmission;(iii) applying a plurality of geometrical delays to the RF channel data for a plurality of candidate SoS values;(iv) constructing a loss metric for each candidate SoS value, the loss metric based on a complex autocorrelation of the time-delayed RF channel data;(v) fitting the loss metric across the plurality of candidate SoS values; and(vi) selecting, as the estimated SoS of the tissue medium, the candidate SoS corresponding to a minimum of the fitted loss metric.

[0057] In an embodiment, the acoustic transmission is a single plane wave transmission.

[0058] In another embodiment, the acoustic transmission is a single focused transmission.

[0059] In another embodiment, the method further comprises constructing the loss metric comprises computing a sum of normalized autocorrelation coefficients (SNAC).

[0060] In another embodiment, the method further comprises ranking the pixels by beamformed intensity and restricting the SNAC calculation to a subset of high- intensity pixels.

[0061] In another embodiment, the method further comprises smoothing the loss metric across candidate SoS values using a spline fit.

[0062] In another embodiment, the method further comprises displaying the estimated SoS in real-time during ultrasound imaging.

[0063] In another embodiment, the estimated SoS is applied to enhance B-mode imaging, Doppler imaging, or tissue characterization.

[0064] In another embodiment, the estimated SoS is computed using GPU acceleration.

[0065] In another embodiment, the method further comprises averaging multiple sequential SoS estimates using a rolling window with trimming of outliers.

[0066] In another embodiment, constructing the loss metric comprises calculating a sum of normalized autocorrelation coefficients (SNAC) across multiple pixels.

[0067] In another embodiment, the method is extended to multiple steered transmissions by independently estimating a speed-of-sound (SoS) for each transmission angle and pooling the estimated SoS values into a single estimate (MTE- IA).

[0068] In another embodiment, the method is extended to multiple steered transmissions by concatenating channel data ensembles from each transmission into a combined dataset and applying the loss metric and fitting process to the combined dataset (MTE-C).

[0069] In another e the medium is a non-biological material, and the estimated SoS is used to evaluate or monitor material composition in a non-destructive testing process.

[0070] 15. The method of claim 1, wherein the medium is meat or livestock tissue, and the estimated SoS is used to evaluate intramuscular fat content, carcass quality, or processed meat quality.an ultrasound transducer configured to generate a single acoustic transmission, the transmission being focused or unfocused, and to receive corresponding echo signals; a processor operatively coupled to the transducer, the processor configured to: acquire pre-beamformed RF channel data from the echo signals; apply geometrical delays to the RF channel data for a plurality of candidate SoS values; compute a loss metric for each candidate SoS value based on a complex autocorrelation of the delayed RF channel data; fit the loss metric across the candidate SoS values; and select the candidate SoS corresponding to a minimum of the fitted loss metric as the estimated SoS; a display configured to present the estimated SoS to a user in real-time.

[0071] In an embodiment, the acoustic transmission is a single plane wave transmission.

[0072] In another embodiment, the acoustic transmission is a focused transmission.

[0073] In another embodiment, the processor is further configured to restrict the loss metric computation to high-intensity pixels.

[0074] In another embodiment, the processor includes a graphics processing unit configured to accelerate autocorrelation and fitting computations.

[0075] In another embodiment, the display is configured to present the estimated SoS alongside a B-mode ultrasound image.

[0076] In another embodiment, the system is configured to operate at a frame rate of at least 20 frames per second.

[0077] In another embodiment, the processor is further configured to average a sequence of SoS estimates using a rolling window and trimming outliers.

[0078] In another embodiment, the processor is configured to calculate SNAC values for each pixel and restrict computation to a subset of high-intensity pixels.

[0079] In another embodiment, the processor applies a smoothing spline to the SNAC- based loss metric before identifying the minimum.

[0080] In another embodiment, the processor is configured to apply a rolling average to consecutive SoS estimates with trimming of highest and lowest values.

[0081] While various embodiments of the invention have been described above with reference to the drawings, it will be appreciated that the scope of the invention is defined by the following claims.REFERENCES

[0082] Ali, R., Brevett, T., Hyun, D., Brickson, L. L., & Dahl, J. J. (2022). Distributed Aberration Correction Techniques Based on Tomographic Sound Speed Estimates. IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, 69(5), 1714- 1726.

[0083] Ali, R., Brevett, T., Zhuang, L., Bendjador, H., Podkowa, A. S., Hsieh, S. S., Simson, W., Sanabria, S. J., Herickhoff, C. D., & Dahl, J. J. (2023). Aberration correction in diagnostic ultrasound: A review of the prior field and current directions. Zeitschrift Fur Medizinische Physik, 33(3), 267-291.

[0084] Anderson, M. E., McKeag, M. S., & Trahey, G. E. (2000). The impact of sound speed errors on medical ultrasound imaging. The Journal of the Acoustical Society of America, 107(6), 3540-3548.

[0085] Bamber, J. C., & Hill, C. R. (1979). Ultrasonic attenuation and propagation speed in mammalian tissues as a function of temperature. Ultrasound in Medicine & Biology, 5(2), 149-157.

[0086] Bezek, C. D., & Goksel, O. (2023). Analytical estimation of beamforming speed- of-sound using transmission geometry. Ultrasonics, 134, 107069.

[0087] Christopher, D. A., Burns, P. N., Hunt, J. W., & Foster, F. S. (1995). The effect of refraction and assumed speeds of sound in tissue and blood on doppler ultrasound blood velocity measurements. Ultrasound in Medicine & Biology, 21(2), 187-201.

[0088] Heriard-Dubreuil, B., Besson, A., Wintzenrieth, F., Cohen-Bacrie, C., & Thiran, J.- P. (2023). Refraction-Based Speed of Sound Estimation in Layered Media: An Angular Approach. IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, 70(6), 486-497.

[0089] Hou, G. Y., Provost, J., Grondin, J., Wang, S., Marquet, F., Bunting, E., & Konofagou, E. E. (2014). Sparse Matrix Beamforming and Image Reconstruction for 2- D HIFU Monitoring Using Harmonic Motion Imaging for Focused Ultrasound (HMIFU) With In Vitro Validation. IEEE Transactions on Medical Imaging, 33(11), 2107-2117.

[0090] Hyun, D., Wiacek, A., Goudarzi, S., Rothlubbers, S., Asif, A., Eickel, K., Eldar, Y. C., Huang, J., Mischi, M., Rivaz, H., Sinden, D., van Sloun, R. J. G., Strohm, H., & Bell, M. A. L. (2021). Deep Learning for Ultrasound Image Formation: CUBDL EvaluationFramework and Open Datasets. IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, 68(12), 3466-3483.

[0091] Jaeger, M., Held, G., Peeters, S., Preisser, S., Grunig, M., & Frenz, M. (2015). Computed Ultrasound Tomography in Echo Mode for Imaging Speed of Sound Using Pulse-Echo Sonography: Proof of Principle. Ultrasound in Medicine & Biology, 41(1), 235-250.

[0092] Jakovljevic, M., Hsieh, S., Ali, R., Chau Loo Kung, G., Hyun, D., & Dahl, J. J. (2018). Local speed of sound estimation in tissue using pulse-echo ultrasound: Model-based approach. The Journal of the Acoustical Society of America, 144(1), 254-266.

[0093] Kamimura, H. A. S., Wu, S.-Y., Grondin, J., Ji, R., Aurup, C., Zheng, W., Heidmann, M., Pouliopoulos, A. N., & Konofagou, E. E. (2021). Real-Time Passive Acoustic Mapping Using Sparse Matrix Multiplication. IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, 68(1), 164-177.

[0094] Krucker, J. F., Fowlkes, J. B., & Carson, P. L. (2004). Sound speed estimation using automatic ultrasound image registration. IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, 51(9), 1095-1106.

[0095] Li, C., Duric, N., Littrup, P., & Huang, L. (2009). In vivo Breast Sound-Speed Imaging with Ultrasound Tomography. Ultrasound in Medicine & Biology, 35(10), 1615-1628.

[0096] Perrot, ., Polichetti, M., Varray, F., & Garcia, D. (2021). So you think you can DAS? A viewpoint on delay-and-sum beamforming. Ultrasonics, 111, 106309.

[0097] Sanabria, S. J., Ozkan, E., Rominger, M., & Goksel, O. (2018). Spatial domain reconstruction for imaging speed-of-sound with pulse-echo ultrasound: Simulation and in vivo study. Physics in Medicine & Biology, 63(21), 215015.

[0098] Sehgal, C. M., Brown, G. M., Bahn, R. C., & Greenleaf, J. F. (1986). Measurement and use of acoustic nonlinearity and sound speed to estimate composition of excised livers. Ultrasound in Medicine & Biology, 12(11), 865-874.

[0099] Stahli, P., Kuriakose, M., Frenz, M., & Jaeger, M. (2020). Improved forward model for quantitative pulse-echo speed-of-sound imaging. Ultrasonics, 108, 106168.

[0100] Stahli, P., Becchetti, C., Korta Martiartu, N., Berzigotti, A., Frenz, M., & Jaeger, M. (2023). First-in-human diagnostic study of hepatic steatosis with computed ultrasound tomography in echo mode. Communications Medicine, 3(1), 1-9.

[0101] Tanter, M., & Fink, M. (2014). Ultrafast imaging in biomedical ultrasound. IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, 61(1), 102-119.

[0102] Telichko, A. ., Ali, R., Brevett, T., Wang, H., Vilches-Moure, J. G., Kumar, S. U., Paulmurugan, R., & Dahl, J. J. (2022). Noninvasive estimation of local speed of sound by pulse-echo ultrasound in a rat model of nonalcoholic fatty liver. Physics in Medicine & Biology, 67(1), 015007.

[0103] Xiao, D., De la Torre, P., & Yu, A. C. H. (2024). Real-Time Speed-of-Sound Estimation In Vivo via Steered Plane Wave Ultrasound. IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, 71(6), 673-686.

[0104] Yiu, B. Y. S., & Yu, A. C. H. (2016). Least-Squares Multi-Angle Doppler Estimators for Plane-Wave Vector Flow Imaging. IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, 63(11), 1733-1744.

[0105] Yoon, C., Lee, Y., Chang, J. H., Song, T., & Yoo, Y. (2011). In vitro estimation of mean sound speed based on minimum average phase variance in medical ultrasound imaging. Ultrasonics, 51(7), 795-802.

Claims

Claims1. A computer-implemented method of estimating a speed-of-sound (SoS) of a tissue medium in real-time, comprising:(i) generating a single acoustic transmission, the transmission being unfocused or focused;(ii) acquiring pre-beamformed radiofrequency (RF) channel data responsive to the transmission;(iii) applying a plurality of geometrical delays to the RF channel data for a plurality of candidate SoS values;(iv) constructing a loss metric for each candidate SoS value, the loss metric based on a complex autocorrelation of the time-delayed RF channel data;(v) fitting the loss metric across the plurality of candidate SoS values; and(vi) selecting, as the estimated SoS of the tissue medium, the candidate SoS corresponding to a minimum of the fitted loss metric.

2. The method of claim 1, wherein the acoustic transmission is a single plane wave transmission.

3. The method of claim 1, wherein the acoustic transmission is a single focused transmission.

4. The method of claim 1, wherein constructing the loss metric comprises computing a sum of normalized autocorrelation coefficients (SNAC).

5. The method of claim 4, further comprising ranking the pixels by beamformed intensity and restricting the SNAC calculation to a subset of high-intensity pixels.

6. The method of claim 1, further comprising smoothing the loss metric across candidate SoS values using a spline fit.

7. The method of claim 1, further comprising displaying the estimated SoS in real-time during ultrasound imaging.

8. The method of claim 1, wherein the estimated SoS is applied to enhance B-mode imaging, Doppler imaging, or tissue characterization.

9. The method of claim 1, wherein the estimated SoS is computed using GPU acceleration.

10. The method of claim 1, further comprising averaging multiple sequential SoS estimates using a rolling window with trimming of outliers.

11. The method of claim 1, wherein constructing the loss metric comprises calculating a sum of normalized autocorrelation coefficients (SNAC) across multiple pixels.

12. The method of claim 1, wherein the method is extended to multiple steered transmissions by independently estimating a speed-of-sound (SoS) for each transmission angle and pooling the estimated SoS values into a single estimate (MTE-IA).

13. The method of claim 1, wherein the method is extended to multiple steered transmissions by concatenating channel data ensembles from each transmission into a combined dataset and applying the loss metric and fitting process to the combined dataset (MTE-C).

14. The method of claim 1, wherein the medium is a non-biological material, and the estimated SoS is used to evaluate or monitor material composition in a non-destructive testing process.

15. The method of claim 1, wherein the medium is meat or livestock tissue, and the estimated SoS is used to evaluate intramuscular fat content, carcass quality, or processed meat quality.

16. The method of claim 1, wherein the medium is a construction material, and the estimated SoS is used to identify curing status, structural integrity, or material composition.

15. A system for real-time estimation of a speed-of-sound (SoS) of a tissue medium, comprising: an ultrasound transducer configured to generate a single acoustic transmission, the transmission being focused or unfocused, and to receive corresponding echo signals; a processor operatively coupled to the transducer, the processor configured to: acquire pre-beamformed RF channel data from the echo signals; apply geometrical delays to the RF channel data for a plurality of candidate SoS values; compute a loss metric for each candidate SoS value based on a complex autocorrelation of the delayed RF channel data; fit the loss metric across the candidate SoS values; and select the candidate SoS corresponding to a minimum of the fitted loss metric as the estimated SoS; a display configured to present the estimated SoS to a user in real-time.

17. The system of claim 16, wherein the acoustic transmission is a single plane wave transmission.

18. The system of claim 16, wherein the acoustic transmission is a focused transmission.

19. The system of claim 16, wherein the processor is further configured to restrict the loss metric computation to high-intensity pixels.

20. The system of claim 16, wherein the processor includes a graphics processing unit configured to accelerate autocorrelation and fitting computations.

21. The system of claim 16, wherein the display is configured to present the estimated SoS alongside a B-mode ultrasound image.

22. The system of claim 16, wherein the system is configured to operate at a frame rate of at least 20 frames per second.

23. The system of claim 16, wherein the processor is further configured to average a sequence of SoS estimates using a rolling window and trimming outliers.

24. The system of claim 16, wherein the processor is configured to calculate SNAC values for each pixel and restrict computation to a subset of high-intensity pixels.

25. The system of claim 16, wherein the processor applies a smoothing spline to the SNAC-based loss metric before identifying the minimum.

26. The system of claim 16, wherein the processor is configured to apply a rolling average to consecutive SoS estimates with trimming of highest and lowest values.

Images