Ultrafast Orthogonal Row–Column Electronic Scanning (uFORCES) With Bias-Switchable Top-Orthogonal-to-Bottom Electrode 2-D Arrays

Top-orthogonal-to-bottom electrode (TOBE) arrays, also known as row–column arrays, have shown great promise as an alternative to fully wired 2-D arrays, owing to a considerable reduction in channels. Novel imaging schemes with bias-switchable TOBE arrays were previously shown to offer promise compared with previous nonbias-switchable row–column imaging schemes and compared with previously developed explososcan methods, however, they required significant coherent compounding. Here, we introduce ultrafast orthogonal row–column electronic scanning (uFORCES), an ultrafast coded synthetic aperture imaging method. Unlike its FORCES predecessor, uFORCES can achieve coherent compounding with only a few transmit events and may, thus, be more robust to tissue motion. We demonstrate through simulations that uFORCES can potentially offer improved resolution compared with the matrix probes having beamformers constrained by the paraxial approximation. Also, unlike current matrix probe technology incorporating microbeamforming, uFORCES with bias-switchable TOBE arrays can achieve ultrafast imaging at thousands of frames per second using only row and column addressing. We also demonstrate the experimental implementation of uFORCES using a fabricated $128 \times 128$ electrostrictive TOBE array on a crossed 25- $\mu \text{m}$ gold wire phantom and a tissue-mimicking phantom. The potential for improved resolution and ultrafast imaging with uFORCES could enable new essential imaging capabilities for clinical and preclinical ultrasound.


I. INTRODUCTION
32 T WO-DIMENSIONAL array transducers have enabled 33 3-D ultrasound imaging but with clinical impact currently 34 limited in part by the image quality. With such 2-D arrays, 35 there exist difficult engineering tradeoffs between system 36 complexity and achievable image quality. Large probes with 37 high-element density would produce high-quality images but 38 with a resulting large number of channels leading to sig-39 nificant interconnect and channel count difficulties. Imple-40 mentation of fully wired arrays is currently prohibitive, with 41 commercial (nonmicrobeamformer) arrays available with only 42 32 × 32 elements, leading to small aperture sizes and poor 43 image quality. Various previous 3-D imaging techniques have 44 been implemented by the mechanical sweeping of a linear 45 or annular transducer but these were not capable of fast 46 volumetric imaging [1]- [3]. A few approaches have been 47 made to reduce the channel count while having a larger 48 aperture size, such as multiplexing and sparsely distributing 49 the active elements, with limited channels but these methods 50 have, thus, far demonstrated sidelobe artifacts that degrade 51 image quality [4]- [6]. Image quality from 2-D arrays has 52 been dramatically improved with the use of microbeamform-53 ing, involving preamplifiers, analog-to-digital converters, and 54 delay-and-sum circuitry implemented as a custom integrated 55 circuit beneath the shadow of each element. 56 In microbeamforming, fine delays are introduced to ele-57 ments before summing in groups, and coarse delays are 58 implemented in the mainframe. Often, microbeamformers 59 implement tilt-only fine-delays as a linear approximation to a 60 quadratic delay profile. These approximations can be a source 61 of image quality degradation, especially when using parallel 62 beamforming to reconstruct a group of adjacent A-scan lines 63 over a wide area, as ideal focal delays are accurate only 64 for one line of sight. As a result, microbeamformer-based 65 matrix probes may not necessarily provide the B-scan image 66 quality, otherwise, found with simpler linear or phased array 67 probes. provided electronic elevational focusing control, electronic 131 scan-plane steering, and 3-D imaging.

132
A significant limitation of previous FORCES and SAFE 133 compounding schemes, however, was the necessity for coher-134 ent compounding over a large number of transmits, which 135 is troublesome in the presence of tissue motion. For a 136 128 ×128 array, FORCES would require motion-free coherent 137 compounding over 128 transmit events, which may not be 138 realistic in many clinical scenarios. Some recent work sought 139 to minimize the number of transmit events for 3-D imag-140 ing using orthogonal plane-wave compounding and nonbias-141 sensitive row-column arrays. However, while enabling fast 142 3-D imaging, significant reconstruction artifacts were present, 143 limiting image quality. Previous work in linear array-based 144 SAI has demonstrated high image quality using sparse trans-145 mission schemes, where the number of transmit events for 146 coherent compounding was limited.

147
In this work, we seek to achieve sparse SAI schemes 148 similar to FORCES, but which require coherent compounding 149 over only a few transmit events. We call our approach ultra-150 FORCES (uFORCES). We demonstrate through simulations 151 that uFORCES can potentially offer improved resolution 152 compared with microbeamformer-based and even fully wired 153 matrix probes constrained by the paraxial approximation 154 in dynamic focusing. Also, unlike current matrix probe 155 technology incorporating microbeamforming, uFORCES with 156 bias-switchable TOBE arrays can achieve ultrafast imaging at 157 thousands of frames per second using only row and column 158 addressing. Using a fabricated 128×128 electrostrictive TOBE 159 array, we also experimentally show the implementation of 160 uFORCES on a crossed 25-μm gold wire phantom. Our work 161 could provide an alternative, and in some cases, improved 162 3-D imaging technology to matrix probe technology, 163 ushering in new opportunities for improved image quality in 164  and 2) uFORCES were implemented with a TOBE array.

173
3) A walking aperture scheme on a fully wired 2-D array 174 (simulating a matrix probe) was implemented for comparison.

175
These imaging schemes are briefly illustrated in Fig. 1. The 176 active aperture is kept the same in all simulations. A walking 177 aperture scheme is selected for the matrix probe as it represents 178 the best possible image quality that could be achieved (in 179 contrast to sector scanning). Additionally, unlike a true matrix 180 probe which implements microbeamformer approximations, 181 we simulate a fully wired array and beamforming constrained 182 to a quadratic delay profile associated with the paraxial 183 approximation.

184
A. FORCES and uFORCES 185 FORCES has successfully been introduced and imple-186 mented in [20] and [21]. In summary, as shown in Fig. 1, 187  . 235 Thus, we apply positive, negative, and positive bias voltages to 236 sparse transmitting columns on columns 3, 8, and 13. Then, 237 we apply a negative bias voltage to all remaining elements 238 as illustrated in the dc pattern #2 in Fig. 2(d). We apply a 239 biasing pattern in this manner for each of the four transmit 240 events. After the complete set of transmit events has been 241 sent, recovered column channel data (inverted when acquired 242 from a negatively biased column) is aperture-decoded using 243 the inverse of the 4 ×4 Hadamard matrix. This then recovers a 244 column data synthetic aperture dataset. As shown in Fig. 2(f), 245 for this example, it recovers channel data as if column 3 first 246 transmitted (with an elevational focus) then data was received 247 on all columns, then column 8 then column 13. A final 248 dataset is recovered which is similar to a plane wave excitation 249 with some of the sparse columns missing (however, it is not 250 used in the beamforming). The synthetic aperture datasets 251 are then reconstructed to form a synthetic aperture image, 252 which is focused on transmission and received everywhere 253 in the scan plane. This is accomplished by beamforming a 254 low-resolution image from each sparse-transmitting element 255 and then coherently compounding the low-resolution images 256 to form a high-resolution image.

257
Steering the scan plane in elevation is also possible when 258 acquiring a volumetric image. In previous sparse SAI work 259 using linear arrays, as few as five sparse transmitting elements 260 had been shown to produce image quality comparable to full 261 SAI [3]. Thus, in what follows, we will use uFORCES with 8, 262 16, or 32 groupings. An eight-transmit uFORCES scheme 263 would recover a synthetic aperture dataset with seven sparse 264 transmitting columns.

265
While our approach requires far fewer transmit events than 266 FORCES, there will be a tradeoff between imaging speed 267 and SNR. The higher the imaging speed, the lower the SNR 268 would be since the effective active aperture with only a few 269 sparse transmitting elements is small. The image quality can 270 be dynamically changed during the imaging by adjusting the 271 number of transmit events where needed.

272
The comparison is conducted in field II [24] with 128 × 273 128 arrays with parameters summarized in Table I. To form a 274 TOBE array, the RF signals of each element on the columns 275 and rows are added up. The effect of the dc bias switching for 276 each pattern was applied to each individual element in field 277 II by alternating the index of the apodization between 1 and 278 −1 denoting positive and negative bias voltages, respectively. 279 Additionally, this apodization is modified with a hamming-280 weighted shape, which is shown to reduce the artifacts caused 281 by side lobes compared with the unity-weighted apodization. 282 The hamming-shaped apodization can potentially be imple-283 mented by tapering the electrodes during the fabrication of 284 the TOBE arrays.  The contrast-detail phantom images of different imaging 319 schemes are compared with each other in terms of CSR, which 320 are calculated using the following expressions [25]: in through-illumination photoacoustic applications [26], [27]. 342 CMUT is another bias-sensitive transducer that uses electro-343 static forces between two clamped plates to generate acoustics 344 which some of their applications in TOBE configurations have 345 been demonstrated recently [8].

346
In this work, a 128 × 128 electrostrictive TOBE array 347 was fabricated to perform uFORCES imaging. This is the 348 largest such TOBE array fabricated to date. The fabrica-349 tion was conducted with steps similar to those previously 350 described [18] for 64 × 64 arrays. As shown in Fig. 3, 351 the transducer is composed of a PMN-PT/epoxy composite 352 material sandwiched in between the top and bottom electrodes 353 which are orthogonal to each other. A quarter-wavelength 354 parylene-C layer was deposited on top as the matching layer 355 and a thick alumina-loaded epoxy on the back serves as a 356 backing layer. Transducer fabrication was performed in the 357 nanoFAB facility at the University of Alberta, Edmonton, 358 AB, Canada. The fabricated array was wire-bonded to a 359 printed circuit board on both the front-and back-sides, which 360 was then connected to an interfacing board connected to 361 our Verasonics Vantage ultrasound platform for testing and 362 imaging. Custom high-voltage biasing electronics were used 363 to apply bias patterns as controlled by the Verasonics system. 364

B. Array Characterization and Testing 365
Prior to Parylene-C deposition, the bias sensitivity of the 366 fabricated transducer was tested by measuring the input 367 impedance for a few different dc biases. The bias-sensitivity 368 for a smaller array was demonstrated previously in [21].  coefficient, k t , as follows [28]:   Fig. 4 illustrates the PSFs of different imaging schemes 429 simulated in field II with a 128 × 128 array. All the images 430 are plotted in a 50-dB dynamic range. The matrix probe 431 walking aperture schemes use narrow dynamically focused 432 reception with an applied 2-D Hanning apodization. We used 433 both narrow and wide transmit beams without any apodiza-434 tions. The narrow beam was created by using the entire 435 active 128 × 128 elements, while for the wide beam, only 436 32 elements in the center were used. Both wide and narrow 437 transmit/receive beams are focused at 25 mm depth. We recon-438 structed images with 501 A-scan lines. However, to compare 439 with sparse-transmitting uFORCES schemes, we considered 440 reducing the number of transmit events. We tested 501, 24, and 441 eight transmit event imaging using these matrix simulations.

442
The images obtained by FORCES and uFORCES meth-443 ods used three elevations (and azimuthal) focusing depths at 444 15, 25, and 35 mm. Images acquired using these different 445 transmit focal depths were then stitched together using a 446 Gaussian-weighted blending algorithm.

452
The calculated lateral and axial resolutions for the PSFs are 453 summarized in Table II. The axial and lateral resolutions are 454 estimated with an error of ±2 μm and ±5 μm, respectively. 455 As can be seen, uFORCES PSFs in Fig. 4(g)    The walking aperture simulation was first done on the 476 phantom with a narrow beam and single focus point at 477 20 mm depth. A total of 601 lines were scanned between 478 −6 and 6 mm lateral distance (x-axis) to form a 2-D image 479 of the phantom. We also simulated a wide transmit beam 480 using the same number of transmit events, and wide-beam 481 excitation using 24 and eight transmit events along with par-482 allel beamforming (Fig. 6). These fewer transmit events were 483 Fig. 7. Simulated comparisons of a contrast detail phantom imaged using (a) TOBE uFORCES and (b) matrix probe wide-beam walking aperture. In both cases, a 10-MHz 128 × 128 lambda pitch array was used but the TOBE array used only row and column addressing. Here, the uFORCES simulation used eight transmit events per focal zone, and stitched results from three elevational focal zones. This required a total of 24 transmit events, with coherent compounding needed over only eight transmit events.  Table III. Visually, uFORCES simulations 499 look crisper owing to improved spatial resolution. Note that the 500 measured CSRs are not better for uFORCES compared with 501 the matrix probe for the middle lesions since this is where the 502 matrix probe is focused on both transmit and receive. However, 503 for the top and bottom lesions, contrast is improved or similar 504 for uFORCES compared with the matrix simulations, and 505 CSRs are similar. The unloaded input impedance of the fabricated transducer 508 with an applied dc bias of 120 V is demonstrated in Fig. 8. 509 The fabricated transducer showed a maximum k t value of 510 ∼0.67 for a voltage of 120 V. As expected, the fabricated 511 bias-sensitive TOBE array shows no piezoelectric effect for a 512 0-V bias voltage while the sensitivity and polarity scale with 513 the bias voltage amplitude and polarity as reported in Fig. 4 Fig. 9 shows results from an immersion transmit test, where 517 a single channel of the array was used to transmit ultrasound, 518 which was reflected from an aluminum plate to be received 519 by the same channel. To this end, a pulser/receiver with 520 an excitation spike voltage of −180 V and a receive gain 521 of 10 dB at a frequency range of 5-20 MHz was used 522 (PANAMETRICS-NDT, 5073PR). The center frequency of the 523 array was measured to be 13.6 MHz with −6-dB bandwidth 524 of 51%.  Our uFORCES simulations demonstrate improved in-plane 551 spatial resolution compared with similar dimension fully wired 552 matrix probes with a walking aperture imaging scheme. 553 We believe this can be explained by two key reasons. is not well achieved without artifacts for f-numbers smaller 563 than unity. In contrast, uFORCES achieves SAI, which is not 564 limited by the paraxial approximation and can achieve fine 565 focusing even for low f-numbers.

566
Elevational focusing with uFORCES is seen to exhibit 567 more beamforming artifacts compared with matrix simulations 568 but the resolution is comparable. uFORCES is limited by 569 unfocused elevational receive elements, even though there 570 is an elevational transmit focus. As such, we used multiple 571 elevational transmit focal zones to improve the depth of field. 572 It should be noted that elevation stitching using multiple 573 transmit focal zones could be achieved without the need for 574 coherent compounding. Thus, even though we used a total of 575 24 transmit events, coherent compounding was needed over 576 only eight such transmit events. This is important as tissue 577 motion can lead to degradation of coherent compounding 578 unless it can be done quickly relative to tissue motion.

579
With current bias tees with a switching time of 300-400 μs, 580 we achieved an imaging rate of >300 fps when using eight-581 transmit uFORCES. With future improvements in bias switch-582 ing electronics, we anticipate thousands of frames per second. 583 Thus, with improved electronics and bias tees, eight-transmit 584 uFORCES with an 8-kHz PRF would result in kHz B-scan 585 imaging rates. In principle matrix probes can transmit wide 586 beams and execute parallel receive focusing to reconstruct 587 many lines at once. However, the fine delays in the microbeam-588 former stage are technically valid for a single-receive line-589 of-sight, and the more parallel beamforming the worse the 590 reconstruction error.

591
In practice, matrix probes will probably not use the walking 592 aperture scheme simulated here. They will likely use all the 593 elements and implement a sector-scanning approach. However, 594 sector-scanning will lead to even more artifacts owing to 595 grating lobes becoming more significant at higher steering 596 angles. The purpose of using a walking aperture scheme here 597 was to compare TOBE uFORCES against the best possible 598 theoretical matrix probe and associated imaging scheme.

599
Imaging advantages over matrix probes are only demon-600 strated in simulation for now. These simulations further 601 included array apodization. This apodization was not yet 602 implemented in array fabrication, but work is underway to 603 do so. Such apodization is important to mitigate edge-wave 604 artifacts and improve imaging point-spread functions.

605
Experiments were conducted with unapodized 128 × 606 128 arrays. Fabrication of these large arrays was found to be 607 highly nontrivial and the tested arrays had 20-25 shorted or 608 dead channels per side, which was a source of some image 609 quality degradation. If a robust fabrication procedure for large 610 TOBE arrays can be developed, these arrays could hold great 611 promise for significant developments in preclinical and clinical 612 imaging applications.

613
Signal-to-noise is degraded using uFORCES compared with 614 FORCES since uFORCES uses a sparse SAI scheme. Strate-615 gies for improving signal-to-noise ratio should be investigated 616 in future work and could include coded excitation schemes, 617 element binning, and so on. Two-dimensional arrays for high-frequency applications 619 do not yet exist commercially. Our technology could 620 achieve this and lead to advances in preclinical 621 ultrasound.

622
Our current experimental results were achieved using a 623 tabletop testbed setup with an integrated water tank. This 624 Fig. 10. Experimental cross-plane uFORCES images of a cross-wire phantom using 32-transmits per elevational focal zone and three such focal zones at 12, 18, and 22 mm depths, (a) XZ plane, (b) YZ plane. These images were obtained by electronically reversing the roles of rows and columns and were obtained without mechanically moving the transducer.   II  IN-PLANE SNR AND RESOLUTION MEASUREMENTS FOR EACH IMAGING SCHEME WITH AND WITHOUT THE NOISE   TABLE III  COMPARISON OF   For the full potential of TOBE arrays to be realized, highly 650 parallelized computing architectures will be needed which 651 may be absent on even state-of-the-art ultrasound platforms. 652 However, the massive explosion of GPU computing accel-653 erated by the deep-learning era will surely prove essential 654 to future high-resolution, massive field-of-view 3-D and 4-D 655 imaging technologies of the future. We envision that TOBE 656 array technology will be an important component of this 657 future wave. Successful realization of uFORCES depends on 658 several practical factors. Ideally, the sensitivity of elements 659 will be uniform but practically, process variations may lead 660 to different responses from different elements. These varia-  Nova Scotia Health Authority, Halifax, as an Affiliated Scientist. His 838 research in high-frequency ultrasound is focused on the development 839 of very high-resolution microfabricated imaging endoscopes for guided 840 surgical applications and low-frequency ultrasound is focused on devel-841 oping miniature highly focused therapeutic transducers for precision 842 tissue ablation. His research interests include piezoelectric transducer 843 design, fabrication, and characterization for both ultrasonic imaging and 844 therapeutic applications and all of the associated electronic hardware 845 required to drive capture, and process the ultrasonic signals.