Improving plane wave ultrasound imaging through real-time beamformation across multiple arrays

Abstract

Ultrasound imaging is a widely used diagnostic tool but has limitations in the imaging of deep lesions or obese patients where the large depth to aperture size ratio (f-number) reduces image quality. Reducing the f-number can improve image quality, and in this work, we combined three commercial arrays to create a large imaging aperture of 100 mm and 384 elements. To maintain the frame rate given the large number of elements, plane wave imaging was implemented with all three arrays transmitting a coherent wavefront. On wire targets at a depth of 100 mm, the lateral resolution is significantly improved; the lateral resolution was 1.27 mm with one array (1/3 of the aperture) and 0.37 mm with the full aperture. After creating virtual receiving elements to fill the inter-array gaps, an autoregressive filter reduced the grating lobes originating from the inter-array gaps by - 5.2 dB. On a calibrated commercial phantom, the extended field-of-view and improved spatial resolution were verified. The large aperture facilitates aberration correction using a singular value decomposition-based beamformer. Finally, after approval of the Stanford Institutional Review Board, the three-array configuration was applied in imaging the liver of a volunteer, validating the potential for enhanced resolution.

Figure 1

Imaging with multiple arrays. (a) Picture of the three-array assembly with its modular 3D-printed stackable manifold. The slight concave geometry minimizes gaps between the arrays. (b) Experimental workflow: each array is connected to a separate Vantage system (part of the 1024 channel volumetric package) where partial GPU beamforming allows efficient processing and real-time display. All the ultrasound systems share the same clock enabling synchronous transmit and receive operation. (c) All arrays are used to transmit plane waves and all arrays are used on receive ultrasound signals. (d) Delay signals sent to generate plane waves taking into account the angle of each array with respect to the central array.

Figure 2

Expanding the imaging aperture effectively improves the resolution as demonstrated by evaluating the point spread function (PSF). (a) Comparison of experimental (top row) and simulated (bottom row) PSF for a 25 µm wire located at a depth of 100 mm. From left to right, the columns correspond to the PSF reconstructed using the central array only, all three arrays, all three arrays with gap compensation and an equivalent fully populated aperture (i.e. no gaps). (b) Lateral cross-section of the PSF displayed in (a). (c) Image of wire targets with known spacing (lateral spacing 1–4 mm with 1 mm increment, axial spacing 1–5 mm with 1 mm increment) located at a depth of 100 mm using the central array only (left) and all three arrays with gap compensation (right). The wires spaced 1 mm laterally cannot be resolved laterally with one array (blue arrow) but are fully separated using three arrays (orange arrows). (d) Lateral cross-section of the targets aligned horizontally displayed in (c). For (a) and (c) the dynamic range is 60 dB.

Figure 3

Extended aperture imaging on a calibrated commercial phantom (CIRS 054GS). (a–d). Images of the phantom utilizing only the central array ((a) above the wire targets; (b) above the hypoechoic cysts) or all three arrays ((c) above the wire targets; (d) above the hypoechoic cysts) for 91 plane waves (− 30° to 30°). (e, f) The expanded view of the deeper set of wire targets (orange rectangle in (a) and (c), located at a depth of 110 mm) shows the significant improvement in term of lateral resolution going from one array (e) to 3 arrays (f). The dynamic range is 60 dB for all images. Note that an agar/glycerol phantom standoff (speed of sound 1540 m/s) is inserted between the array assembly and the top surface of the phantom. (g) Comparison of the lateral resolution (full-width at half maximum) as a function of depth for the set of wires located at the center of (a) and (c).

Figure 4

Quantification of contrast on the hypoechoic cyst located at 80 mm as a function of the number of plane waves. (a–c). Expanded view of the hypoechoic region located at 80 mm (green rectangle in Fig. 3 (b) and (d)) going from one array (a) to three arrays (b) and after applying the SVD beamformer (c). The ROIs indicate the inside and outside areas selected to calculate the contrast metrics. The dynamic range is 60 dB for all images. (d) The autocorrelation of the speckle as a function of the lateral distance confirms the change in speckle aspect utilizing three arrays with reduced lateral correlation distance. (e) The contrast ratio as a function of the number of angles shows a slight degradation going from one to three arrays that is partially compensated by the SVD beamformer. (f, g) The contrast-to-noise ratio (f) and generalized contrast-to-noise ratio (g) as a function of the number of angles are very similar between one and three arrays, especially for a higher number of angles. A small degradation is seen for three arrays for a small number of angles but is compensated when applying the SVD beamformer.

Figure 5

Evaluation of aberration correction on the array assembly with the SVD beamformer on the calibrated commercial phantom (CIRS 054GS). (a) Delay-and-sum (DAS) image of the phantom with proper speed of sound matching between the arrays and the top flat surface of the phantom (the yellow trapezoidal contour highlights the agar/glycerol standoff). (b) A numerical aberration is applied on the receive data resulting in image degradation. (c) The application of the SVD beamformer retrieves the original image. (d) Comparison of the aberration estimated from the SVD processing (red) to the numerical aberration applied to the receive data (blue) as a function of the angle number. The isoplanatic patch used for the estimation is indicated by the orange square in (c). (e) The standoff is replaced with a water/ethanol solution (35/65 volume ratio) reducing the speed of sound to 1450 m/s and degrading the DAS image. (f) Image degradation is reduced after applying the SVD beamformer. (g) Aberration estimated from the SVD processing as a function of the angle number. The isoplanatic patch used for the estimation is indicated by the square in (f). The dynamic range is 60 dB for all images.

Figure 6

Quantification of contrast on the hypoechoic cyst located at 80 mm (see Fig. 5) as a function of the number of plane waves when aberrations are introduced. (a–e) Expanded view of the cyst without aberration (a), after introducing a numerical aberration before (b) and after (c) applying the SVD beamformer, after introducing an experimental aberrating layer before (d) and after (e) applying the SVD beamformer. The dynamic range is 60 dB for all images. (f–h). The contrast ratio (f), contrast-to-noise ratio (g) and generalized contrast-to-noise ratio (h) as a function of the number of angles shows a clear degradation in the presence of aberration.

Figure 7

Imaging of liver on a volunteer using the central array (a, g), all three arrays (b, h) and after applying the SVD beamformer (c, i). The close up view indicated in (a) and (g) by a yellow square is displayed for all three cases in (d) and (j) (one array), (e) and (k) (three arrays) and (f) and (l) (three arrays + SVD). The dynamic range for all the images is 60 dB. The circular regions of interest (ROI) indicates where the contrast metrics were evaluated.