HTML
-
A right-angle prism made of UV fused silica (PS611, Thorlabs, Inc.; 2203 kg/m3 density, 73.6 GPa Young's modulus) was chosen as the acoustic ER12. A miniature ultrasonic transducer (XMS-310, Olympus, Inc.; 10 MHz central frequency, 2 mm element size) was placed at a corner of the prism to break the symmetry. A two-channel digitizer (ATS9350, AlazarTech, Inc.; sampling rate of up to 100 MS/s) recorded the encoded PA signals. A 532-nm wavelength pulsed laser (INNOSAB IS8II-E, Edgewave GmbH; 2-kHz pulse repetition rate and 5-ns pulse width) was used for optical excitation. The laser beam was filtered and expanded by a pinhole and two lenses.
Prior to the multifocal measurement, a training step was performed to quantify the impulse response for each pixel across the FOV. A focusing lens (LA1509, Thorlabs, Inc.; 25.4-mm diameter and 100-mm focal length) was used to focus the laser beam to a 5-μm spot. The laser radiant exposure (or fluence) was 50 mJ/cm2. Via the PA effect18, 19, PA waves were generated by focused laser excitation and propagated through the ER. The focal diameter and pulse width of the laser beam were much narrower than the central wavelength and the reciprocal of the bandwidth of the ultrasonic transducer, respectively. Consequently, the PA wave input to the PAMER system could be approximated as a spatiotemporal delta function, and the detected signals quantified the impulse response of the linear system for the excitation position. The focused laser spot was raster scanned over the entire FOV—with a customized scanner consisting of two motorized translation stages (PLS-85, PI GmbH & Co.)—to record the point-by-point impulse responses.
After the training step, the focusing lens was replaced with a microlens array (64–479, Edmund Optics, 500-μm pitch, 1.2° divergence angle) to generate multiple optical foci. The working distance of a microlens array is relatively short, preventing us from focusing the optical foci onto the sample after passing through the ER. Therefore, a relay lens (272EN II, Tamron, 0.29 m minimum focus distance, 1:1 maximum magnification ratio) was used to increase the working distance while preserving the size of the optical focal spots. The laser radiant exposure at each optical focal spot was maintained at ≤20 mJ/cm2. PA signals generated from the multiple optical foci were then detected by the single-element ultrasonic transducer.
-
The PA signals propagating through the ER can be expressed as a linear combination of the impulse responses from all the illuminated pixels:
$$ s\left( t \right) = \mathop {\sum}\limits_{i = 1}^{N_p} {k_i} \left( t \right)P_i $$ (1) where s(t) is the PA signal detected through the ER, i is the pixel index, Np is the total number of pixels, ki(t) is the normalized impulse response, and Pi is the local PA amplitude at the ith pixel. Equation (1) can be recast in matrix form by discretizing time t according to the Nyquist criterion:
$$ {\bf{s}} = K{\bf{P}} $$ (2) where $K = \left[ {k_1, \ldots ,k_{N_p}} \right]$ is the system matrix and P is the RMS PA amplitude image12. A two-step iterative shrinkage/thresholding (TwIST) algorithm20 was implemented to solve Eq. (2) for P as a minimizer of the objective function:
$$ \widehat {\bf{P}} = {\arg} \,{\mathop {\min}\limits_{\bf{P}}}\left\| {{\bf{s}} - K{\bf{P}}} \right\|^2 + 2\lambda {\mathrm{{\Phi}}}_{{\mathrm{TV}}}({\bf{P}}) $$ (3) Here, ΦTV(P) is the total variation regularization term, and λ is the regularization parameter.
To generate a partial MFOR image, each reconstructed multifocal image was localized based on the true positions of the optical foci. The reconstructed images were digitally upsampled using the imresize function in MATLAB, the maximum amplitude of the pixels within each optical focal spot size (i.e., 13/2-μm radius around the localized centre position) was determined, and all pixels within the optical focal spot were set to that maximum value. Pixels outside the optical foci were zeroed out. Finally, all the partial MFOR images were summed to generate a 2D MFOR-PAMER image. During the multifocal measurement, only a distance equal to the pitch of the microlens array needs to be scanned to form a 2D MFOR-PAMER image. Therefore, the scanning time can be shortened by a factor equal to the number of microlens elements on the array to cover the same FOV.
-
A training dataset of 500 × 500 pixels with a stepsize of 20 μm was first acquired by covering the imaging surface of the ER with black acrylic paint uniformly. To synthesize the multifocal measurement, we set Pi = 1 at the pixel positions where optical foci were generated by a simulated 2D microlens array and Pi = 0 at the other positions. The impulse responses from the training dataset for pixels under Pi = 1 were added up in the synthetic multifocal measurement. A zero-mean Gaussian random vector representing white noise was added to the synthesized signals. To quantify the relationship between the minimum separation pitch and the central wavelength, we used three ultrasonic transducers with similar physical parameters at central frequencies of 5 MHz (VP-0.5–5 MHz, CTS Electronics, Inc.; 5 MHz central frequency, 0.5 mm element size), 10 MHz (XMS-310, Olympus, Inc.; 10 MHz central frequency, 2 mm element size) and 20 MHz (VP-0.5–20 MHz, CTS Electronics, Inc.; 20 MHz central frequency, 0.5 mm element size).
-
For the measurement of the AR resolution, a training dataset of 100 × 50 points with a stepsize of 15 μm was first acquired by covering the imaging surface of the ER with a uniform layer of black acrylic paint (100× averaging, acquisition time = 5 min). Two 5-μm-diameter laser spots were simultaneously shone onto the black paint. While one beam was held stationary, the other beam was translated linearly away from the first during the PA measurements. For the measurement of the OR, a training dataset of 80 × 40 points with a stepsize of 2 μm was first acquired by covering the ER imaging surface with a thin metal blade painted with black acrylic paint uniformly (100× averaging, acquisition time = 4 min). After the training step, the metal blade was repositioned to have its edge resting in the middle of the FOV. A customized water tank was attached to the sample to reduce reconstruction degradation after the repositioning of the blade.
-
A piece of transparency film was cut to 25 mm × 25 mm and painted with black ink on one side for the training step. Since the imaged object is effectively a part of the ER system response, the leaf skeleton was attached to the film with ultrasonic gel to facilitate acoustic coupling. After the training step, the painted film was replaced with an unpainted film to image the leaf skeleton with the microlens array. A training dataset of 500 × 500 points with a stepsize of 20 μm was first acquired by covering the ER imaging surface with a thin film (~250 μm, similar to the thickness of the leaf skeleton phantom) painted with black acrylic paint uniformly (10× averaging, acquisition time = 30 min). After the training step, the thin film was replaced with the leaf skeleton phantom. A customized water tank was attached to the film and the phantom during the experiment to reduce reconstruction degradation after swapping the imaging samples. The microlens array was raster scanned with a 500 μm × 500 μm range and a stepsize of 20 μm to image the entire FOV.
-
Female ND4 Swiss Webster mice (Envigo; 18–20 g and 6–8 weeks) were used for the animal study. The laboratory animal protocols were approved by the Institutional Animal Care and Use Committee of the California Institute of Technology. The mouse was anaesthetized with 5% vaporized isoflurane mixed with air to induce anaesthesia and then transferred to a customized animal mount allowing the mouse ear to be laid flat on the imaging face of the ER. The mouse was anaesthetized with a continuous supply of 1.5% vaporized isoflurane during the experiment. A training dataset of 500 × 500 points with a stepsize of 20 μm was first acquired by covering the ER imaging surface with a thin film (~250 μm, similar to the thickness of the mouse ear) painted with black acrylic paint uniformly (10× averaging, acquisition time = 30 min). After the training step, the thin film was replaced with the mouse ear. A customized water tank was attached to the film and the mouse ear during the experiment to reduce reconstruction degradation after swapping the imaging samples. The surface optical fluence at each optical focal spot through the microlens array was maintained at ≤20 mJ/cm2 to comply with the ANSI safety limit per laser pulse21.
-
To satisfy the Nyquist criterion, the theoretical AR training stepsize needs to be < 1/2 the AR imaging resolution (~100 μm for the current setup, as shown in Fig. 3). If we utilize the same laser at 2 kHz, the scanning stepsize will be ~70 μm to cover a 10 × 10 mm2 FOV in 10 s.