A perspective on high-frequency ultrasound for medical applications

Abstract

High-frequency ultrasound (HFU, > 15 MHz) is a rapidly developing field. HFU is currently used and investigated for ophthalmologic, dermatologic, intravascular, and small-animal imaging. HFU offers a non-invasive means to investigate tissue at the microscopic level with resolutions often better than 100 μm. However, fine resolution is only obtained over the limited depth-of-field (~1 mm) of single-element spherically-focused transducers typically used for HFU applications. Another limitation is penetration depth because most biological tissues have large attenuation at high frequencies. In this study, two 5-element annular arrays with center frequencies of 17 and 34 MHz were fabricated and methods were developed to obtain images with increased penetration depth and depth-of-field. These methods were used in ophthalmologic and small-animal imaging studies. Improved blood sensitivity was obtained when a phantom mimicking a vitreous hemorrhage was imaged. Central-nervous systems of 12.5-day-old mouse embryos were imaged in utero and in three dimensions for the first time.

1. Introduction

High-frequency (i.e., f >15 MHz) ultrasound (HFU) is under considerable investigation because the short wavelengths (e.g., 100 μm at 20 MHz) and small focal-zone beam diameters of HFU provide fine-resolution images. Studies have demonstrated the unique ability of HFU systems to image shallow or low-attenuation tissues for biomedical applications. For example, HFU has been successful for small-animal [1], ocular [2], intravascular [3], and dermatological imaging [4]. However, HFU penetration depth is reduced because of frequency-dependent attenuation. Furthermore, most current HFU images have a small depth-of-field (DOF) because low F-number transducers are employed to improve cross-range resolution. Therefore, typical HFU images have fine resolution only at shallow depths and over a limited DOF.

In this study, two 5-element annular arrays with center frequencies of 17 and 34 MHz were fabricated. Annular arrays represent an intermediate step between single-element transducers and linear arrays. Reconstructed images using a synthetic-focusing algorithm have increased DOF (>4 mm), but the annular array must be mechanically scanned to form an image [5]. The synthetic-focusing algorithm was complemented by using chirp-coded signals to excite the annular-array transducer [6]. Chirps allowed the received-signal SNR to increase by using a matched filter on receive.

The annular-array, chirp-coded imaging methods were used in ophthalmologic and small-animal-imaging studies. An ex-vivo bovine eye was imaged with the 17-MHz annular array. Also, a phantom that mimicked a vitreous hemorrhage was designed and imaged with the 34-MHz array. Finally, the central-nervous systems of 12.5-day-old mouse embryos were imaged in utero and reconstructed in three dimensions.

2. Methods

2.1. Annular-array design

The fabrication of the annular arrays was described in previous studies [5]. The 34-MHz arrays consisted of a 9-μm thick copolymer membrane (Ktech Corp., Albuquerque, NM) bonded to a single sided, copper-clad polyimide (CCP) film (RFlex 1000L810, Rogers Corp., Chandler, AZ). An array pattern was etched onto the CCP using standard printed-circuit-board etching techniques. The 17-MHz array was designed with the same protocol but a 25-μm membrane was used [6]. All the arrays used in these studies had five equal-area annuli. The 34-MHz array had a radius of curvature of 12 mm and a total aperture of 6 mm. The 17-MHz array had a radius of curvature of 31 mm and a total aperture of 10 mm. After impedance matching, all the elements of both transducers had about 35% 6-dB fractional bandwidth.

2.2. Chirp-coded signals

In this study, linear chirp-coded excitations were used to excite each array element. A linear chirp is a coded signal that linearly spans a frequency bandwidth$B=f_2-f_1$, where $f_2$ and $f_1$ are the starting and ending frequencies, respectively. If the chirp sweeps from $f_1$ to $f_2$ over a time, T, then the chirp-coded excitation is described by

$$s(t)=w(t)\text{cos}(2\pi f_{1}t+\pi b{t}^2),$$ (1)

where w(t) is a 9%-Tukey windowing function that vanishes outside of [0, T] and b is the sweep rate equal to (f2 – f1)/T. The other parameters were different for the two annular arrays and are shown in Table 1.

Table 1.

Chirp parameters

Center frequency	F 1	f 2	T
17 MHz	6.5 MHz	32 MHz	8 μs
34 MHz	15 MHz	65 MHz	4 μs

2.3. Data acquisition and image formation

The experimental system was described previously in detail [5, 6]. Briefly, data for a B-mode image were acquired by making five sequential lateral passes across a sample. On each pass, one of the array elements was excited with a mono-cycle pulse or a chirp coded-excitation, and receive signals were digitized on all five array elements at 200 or 400 MHz (depending on the array center-frequency). Using this method, all 25 transmit/receive data pairs were captured. The experimental system was completely automated, and acquiring a full set of data required about 4 s.

For the datasets acquired using chirp-coded excitation, each received echo signal was compressed before synthetic focusing. Pulse compression is meant to restore satisfactory axial resolution before forming the image. Pulse compression was done by linearly filtering each transmit/receive pair with the compression filter. In this study, the compression filter was taken to be the time-reversed chirp weighted by a Dolph-Chebyshev window. No pulse compression was necessary when the data were acquired using a conventional pulsing method.

The radio-frequency (RF) data were then post-processed with a synthetic-focusing algorithm to improve the image DOF and lateral resolution. To focus at a depth, d, appropriate digital time shifts were applied to each RF transmit/receive data line [5], and the resulting 25 lines were summed to create a single, focused line. The summed line was then windowed to extract only a region of data around the focal depth d. This process was repeated for n focal zones, and a final composite image was formed by combining all of the focal zone segments. The spacing between adjacent focal zones was 0.2 and 0.4 mm for the 34- and 17-MHz annular arrays, respectively. Direct summation of the 25 lines simulated a single-element transducer with the same focal properties as the annular array. Direct summation was used in some of the experiments to compare the performance of the annular array to a single-element transducer with the same focal properties.

3. Results

3.1. Axial and lateral resolutions

To determine how the chirp-coded excitation method affected image resolution, a 12-μm diameter tungsten wire was scanned at depths ranging from 20 to 41 and 8 to 18 mm, for the 17- and 34-MHz arrays respectively. Images were formed at each depth and the 6-dB axial and lateral resolutions were measured for different imaging methods.

Figures 1a and 1c indicate that for both annular-array transducers, the axial resolution obtained using the chirp-imaging method is about the same as that obtained using conventional high-frequency pulsers (i.e., Avtech and Panametrics). Figures 1b and 1d display the −6-dB lateral resolution as a function of depth for both annular arrays, and demonstrate that the lateral resolution achieved by all imaging methods was essentially the same (as theory predicts). The lateral resolution worsens (i.e., increases) with depth because, as expected, the synthetic-focusing algorithm increases the effective transducer F-number as the focal depth increases with a fixed aperture.

Fig.1.

Axial (a and c) and lateral (b and d) −6-dB resolutions obtained as a function of depth for both annular arrays using different excitation methods.

3.2. Penetration depth and SNR

Our second experiment investigated whether coded excitation increases SNR and penetration depth compared to what can be achieved with conventional imaging. For this experiment, a tissue-mimicking phantom was scanned using chirp and Avtech excitation. The tissue-mimicking phantom (ATS Laboratories, Bridgeport, CT) contained 10-μm diameter glass beads (8×10⁶ beads/cm³⁾ and had an attenuation of 0.5 dB/cm/MHz near 6 MHz. Following image formation, the penetration depth into the phantom and the SNR were estimated (Table 2).

Table 2.

Penetration depth and SNR based on the tissue-mimicking phantom images

Center frequency	Method	SNR (dB)	Penetration depth (mm)
17 MHz	Avtech	46.9	11.8
17 MHz	Chirp	58.8	15.5
34 MHz	Avtech	59.3	3.7
34 MHz	Chirp	73.0	7.1

Results indicate that for both annular-array transducers SNR and penetration depths significantly increased when the chirp-imaging method was used. SNR increased by more than 11 dB and penetration depth by more than 3.4 mm.

3.3. Bovine-eye imaging

Two ultrasound images of a bovine eye are displayed in Fig. 2. In both images, the geometric focus of the transducer (31 mm) is just beyond the lens. Both images were obtained using the 17-MHz annular array. To obtain Fig. 2a, a conventional pulsing method (i.e., Avtech) was used. This image demonstrates that the anterior and posterior segments were in sharp focus simultaneously because of the extended DOF allowed by the array. A detachment of the corneal epithelium (ED) was clearly depicted. Similarly, the anterior chamber (AC) was sharply visualized.

Fig.2.

Ex-vivo images of a bovine eye obtained with the 17-MHz annular array: (a) Avtech, 60-dB dynamic range; (b) chirp, 75-dB dynamic range.

Improvement in image quality was observed in the chirp image (Fig. 2b); not only was the whole eye in sharp focus, but the increase in SNR (+17 dB compared with Fig. 2a) allowed visualization of structures that previously were hidden by noise. The lens circumference was fully visible, and scattering from the vitreous was seen proximal to the retina. Fig. 2b demonstrates the strength of combining annular-array imaging with coded excitation. The annular array allows for the creation of an image with an extended DOF, and the chirp allows for deeper penetration and for the display of weaker specular reflectors. The practical implication was that annular array imaging with synthetic focusing and chirp excitation could be diagnostically significant.

3.4. Blood phantom

A phantom containing fresh rabbit blood suspended in agar was fabricated to model a vitreous hemorrhage. To make the phantom, 15 g of agar was diluted in 1 L of normal saline solution. After stirring, the solution was heated in a microwave oven until the agar was dissolved, which took approximately one minute. Droplets of fresh rabbit blood were injected with a needle into each container while the agar was still warm and not fully cured. The agar phantom had low attenuation and its acoustic properties were similar to those of vitreous. Also, the blood droplets constituted a low-contrast (i.e., weak-backscatter) target because they were small and had an acoustic impedance value similar to that of agar. Data were collected with the 34-MHz annular array using Avtech and chirp excitations. Imaging was performed with the front surface of the phantom placed 8 mm away from the transducer surface (i.e., 4 mm before the natural focal depth of the transducer).

Figures 3a and 3b display the resulting images using Avtech and chirp excitation, respectively. For comparison, both images are displayed with a 45-dB dynamic range. The Avtech excitation case (Fig. 3a) had modest SNR (i.e., 33.2 dB) because the blood droplets were small and were weak scatterers. At depths ranging from 10 to 12 mm, the blood was visualized, but beyond 12 mm, the blood was not as visible because of the noisy background and of the weak blood backscatter. The chirp excitation case (Fig. 3b) had a greatly improved SNR (i.e., 49.0 dB, +15.8 dB). This improved SNR led to an image with greater contrast, and distinct blood droplets were visualized significantly deeper (i.e., > 16 mm) into the phantom (Fig. 3b).

Fig.3.

Images of a blood phantom obtained with the 34-MHz annular array using the Avtech pulser (a) and the chirp-imaging method (b). The displayed dynamic range is 45 dB.

3.5. Mouse-embryo imaging

Mouse embryos were imaged with the 34-MHz annular-array transducer at embryonic day (E) 12.5, where E0.5 was defined as noon of the day a vaginal plug was found after overnight mating. In these studies, the pregnant mouse was euthanized humanely by cervical dislocation immediately before image acquisition to eliminate breathing motion. With the mouse placed in a supine position and its four limbs taped to an imaging stage, the stomach was shaved and a Petri dish with a circularly cut 25-mm imaging window was placed over the stomach, filled with isotonic saline, and secured. For imaging, the annular-array transducer was lowered into the saline-filled Petri dish.

Three-dimensional data sets were acquired with the geometric focus of the transducer placed superficial to an embryo (Figs. 4a and 4b, yellow arrows) to simulate the most challenging situation to visualize brain ventricles. In this configuration, the ventricles were far from the natural focus of the annular array and deep (~6 mm) into attenuating tissue. Figures 4a and 4b show the same representative slice from the stack of reconstructed images using Avtech and chirp excitation, respectively. On both images, synthetic focusing increased the DOF enough to resolve the entire contour of the embryonic head. On Figs. 4a and 4b, the yellow and purple regions were the automatically segmented brain ventricles. The pink region was the manually segmented head of the embryo. The increase in SNR allowed by the chirp-imaging method permitted better resolution of the complete two-dimensional extent of the brain ventricles. In particular, more artifacts were present in the Avtech image with segmented brain ventricles outside the embryonic head.

Fig.4.

a) and b) Representative B-mode images and brain-ventricle segmentation results from the Avtech and chirp datasets, respectively. The segmentation of the embryonic head is also shown (pink contour). White bar represents 1 mm. Surface segmentation of embryonic brain ventricles from the Avtech (c) and chirp dataset (d). The chirp dataset shows the two lateral ventricles (lv) and the third (3v) and fourth (4v) ventricles. The surface rendering of the embryonic head (e and f) visualizes anatomical features such as the ear (e), limb bud (l), and the eye (i). Registration with the embryonic head reveals errors in the volume reconstruction of the brain ventricles for both datasets (arrows in e and f).

Three-dimensional renderings of the embryonic brain ventricles are visualized for Avtech (Fig. 4c) and chirp (Fig. 4d) excitation. The images reconstructed from the Avtech dataset showed significant artifacts (indicated by the arrows in Fig. 4c). However, all four ventricles were clearly visible (denoted by lv, 3v, and 4v in Fig. 4d) without obvious artifacts in the chirp-based reconstruction in Fig. 4b. Following segmentation and rendering, total brain cavity volumes were estimated. The Avtech-based volume estimate (11.4 mm³) was 80% larger than the chirp-based volume estimate (6.2 mm³).

Segmentation of the embryonic head showed several anatomical features such as the ear, eye, and limb bud (denoted by e, i, and l on Fig. 4f). Co-registration of the embryonic head with the brain ventricle volumes allowed visualizing the spatial extent and orientation of the ventricles with respect to the head. In particular, co-registration of these volumes revealed errors made by the segmentation algorithm because of the lack of contrast between the brain cavity and the surrounding tissue. One of the errors resulted in the placing of the location of brain cavities outside the head of the embryo, as shown by the arrows in Fig. 4e and 4f. These images also revealed that the greatest errors were made by the Avtech-based ventricle reconstruction.

4. Conclusions

The increased SNR, DOF, and penetration depth allowed by chirp-coded excitation permitted the formation of better HFU images for several medical applications.

The in vitro bovine-eye images demonstrated that synthetically-focused, chirp-derived images contained additional diagnostically significant information for detecting and assessing ocular pathologies when compared to current state-of-the-art methods.

Chirp imaging also allowed the in utero visualization of the brain-ventricle system with satisfactory contrast. Thus, 3D segmentation of the brain-ventricle system from the chirp dataset was possible with minor errors whereas significant errors were present in the Avtech-based segmentation. The superior robustness of the chirp-imaging method will enable translational studies of normal and abnormal development of the embryonic nervous system.

In conclusion, chirp-coded excitation represents a valuable method for improving HFU images. Implementation in clinical instruments could be done in a manner that is transparent to the end user. The chirp images would have the same format and properties as familiar, conventional B-mode images, except for a noticeably improved SNR. The salient advantage of such images is that their improved quality (in terms of DOF and SNR) can help detection and evaluation of disease conditions when visualization of fine tissue structures is necessary.