2.1.1.

Phase (i): module specification verification

2.1.2.

Phase (ii): image quality characteristic assessment

2.2.1.

Optical and acoustic material properties

The relevant optical and acoustic properties of the materials that the test objects are built from were characterized. Acoustic sound speeds and attenuation were measured with US transmission measurements following the method described in Ref. 51. The optical absorption of non-scattering media was calculated from transmission measurements in a spectrophotometer following the Lambert–Beer law. Optical properties of scattering media were calculated from transmission and reflection measurements in the same spectrophotometer with the inverse adding doubling method.⁵⁵ The most relevant acoustic and optical material properties are included in the test object descriptions in the results section. Additional material properties can be found in Fig. S4 in the Supplementary Material.

2.2.2.

Photoacoustic imaging

The test objects were imaged with the Twente PAM 2³¹ to demonstrate their PA appearance and show their potential use. The PAM 2 system is a tomographic system that is equipped with a dual-head laser, emitting at 755 and 1064 nm. A bottom beam illuminates the breast from the nipple side and nine fiber bundles illuminate the breast from the sides. The system contains a 1-MHz-centered hemispherical detection aperture that consists of 12 arc-shaped arrays. The arrays are radially positioned and rotate around their central coordinate during a measurement to obtain multiple projections. PA images were reconstructed from the recorded time traces with an iterative back-projection algorithm. A detailed description of the system and the reconstruction algorithms can be found in Refs. 31 and 54. The well-known test object shapes allowed one to employ a two-layer SOS model in the reconstruction, where one SOS was assigned to the coupling water and one to the test object volume. The SOS assigned to the water was determined by the water temperature that was measured for each measurement. The SOS assigned to the test object volume varied per test object and reconstruction and is specified with the images in the next section.

3.1.1.

PA point source

A single subresolution PA point source was developed, which can function as an object for early system testing in phase (i) to validate the data acquisition and PA reconstruction algorithms. Its second use lies in the measurement of the detection bandwidth and relative sensitivity of all the US detection elements. As a third use, the object can also be used in monitoring the detector performance over time. The basis of this test object is a stainless steel microsphere with a 310- to $360\text{-}\mu \mathrm{m}$ diameter (Cospheric LLC, United States). A microscope image of these beads is included in Fig. S1 in the Supplementary Material. The beads were dyed by dipping it in India ink (Royal Talens, The Netherlands) and letting it dry at room temperature. The inset in Fig. 2(a) shows a microscope image of a coated bead.

A thin ($180\mu \mathrm{m}$) circular PVC sheet (Krumbeck Kunststoffverarbeitung GmbH, Stadtlohn, Germany) was vacuum formed over an 8.7-mm-thick silicone disk. The bead was then glued to the center of the PVC sheet using a droplet of superglue (Loctite 406, United States). This enabled positioning of the microsphere at the central coordinate of the hemispherical detection aperture of the PAM 2 system, equidistant to all detectors while using the standard cup mounting system. The PVC sheet is known to minimally disturb the PA measurement due to its high optical and acoustic transparency.⁵⁴

3.1.2.

Black cups

As a second test object, we developed breast supporting cups in eight different sizes from black PVC. The starting material was a $1\pm 0.1\text{-}\mathrm{mm}$-thick black PVC-U sheet (SIMONA AG, Germany), which was vacuum-formed using a vacuum shaping machine (Formech 300, Formech International Limited, UK). The same moulds that are normally used to produce the transparent breast supporting cups⁵⁴ for the PAM 2 system were used. The reconstructed pressures from PA measurements on these objects will be a representation of the illumination profile on the breast surface. With this knowledge, the OI models for QPA can be made cup-specific.

3.1.3.

PA spatial resolution object

The PA spatial resolution is an IQC that can potentially be optimized by changes in the measurement sequence or improvements made in reconstruction algorithms. A test object containing 277 subresolution India ink coated stainless steel microspheres spread out over the imaging volume was developed to measure the spatial resolution in all three dimensions and through the entire imaging volume. The microspheres were held in a polyvinyl plastisol (PVCP) matrix with a $1403\pm 3\mathrm{m}/\mathrm{s}$ sound speed at 1 MHz. No scatterers or absorbers were added to the PVCP to keep it optically and acoustically semitransparent. A Nylon (PA2200) mould in the shape of the largest breast supporting cup was 3D printed with the Formiga P101 printer (EOS, Germany) (see Fig. S2 in the Supplementary Material). Nylon was chosen as it is known to withstand the high temperature of the liquid PVCP ($\pm 180°\mathrm{C}$). The large volume of this mould (1.2 l) and the fact that large volumes of PVCP tend to get overheated and burn during the polymerization process led us to build this phantom layer-by-layer. For each layer, a volume of PVCP (LureFlex firm, LureFactors, UK) was measured and 1 v/v% heat stabilizer (M-F Manufacturing, Texas, USA) was added to make it better resistant to heat. The heat stabilizer was mixed through the PVCP for 5 min with a magnetic stirrer. To assure an equal layer thickness, the PVCP volumes ranged from 50 to 400 ml, where the smaller volumes were used for the first layers in the bulge of the cup mould. The opaque PVCP mixtures were degassed for 5 min and heated in the microwave at 270 W in steps of 1 min until the mixture transformed into a transparent viscous liquid. The mixture was slightly swirled but not stirred to prevent air bubbles from forming. Subsequently, the PVCP was poured into the mould, leaving the not fully polymerized PVCP in the beaker. Directly after pouring, several stainless steel beads were distributed on the surface [see Fig. S2(b) in the Supplementary Material]. This process was repeated 11 times to fill the mould to the top. The 11th layer acts as a shutoff layer to protect the beads in the 10th layer. The object was placed in a size 8 breast supporting cup to install it in the imager.

3.1.4.

Speed of sound object

A test object containing materials with different sound speeds was developed to test the ability of the US transmission modality in calculating SOS maps. The inclusion of PA absorbers in this test object allows one to investigate improvements in the PA spatial resolution when the object’s sound speed, known a priori or measured with US transmission measurements, is used in the PA reconstruction.^14,48,56

The object was once again given the shape of the largest breast supporting cup and includes sections that have different sound speeds. The material interfaces were given a gradually curved shape to minimize diffraction artifacts. The test object was built from two thin transparent ($180\mu \mathrm{m}$) vacuum-formed PVC cups (Krumbeck Kunststoffverarbeitung GmbH, Stadtlohn, Germany) that were put together. The outer cup is a size 8 breast supporting cup, and the therein placed cup is a smaller 3D ellipsoid and is positioned off-center. The same India ink coated microspheres as used in the other test objects were glued to the outside of the ellipsoid cup [Fig. 3(a)]. The volume between the two cups was filled with degassed traditional olive oil (Carbonell, Spain) with a $1459\pm 3\mathrm{m}/\mathrm{s}$ sound speed, forming the outer layer. Two bolted metal rings were used to clamp the two PVC sheets and a rubber ring together to avoid leakages of the oil. The hemispherical cup was then filled with polyacrylamide gel (PAG) with a $1512\pm 3\mathrm{m}/\mathrm{s}$ sound speed that was prepared by mixing 82.08 v/v% MiliQ, 17.5 v/v% acrylamide-bisacrylamide (VWR, USA, catalog number SERA10679.02), and 0.14 w/v% ammonium persulfate (Sigma Aldrich, USA, catalog number 248614).⁵⁷ Two modeling balloons (Modelleerballon, Art. No. 10057434, Fun & Feest, The Netherlands) filled with water were used as placeholders for channels [Fig. 3(b)] and were positioned in the mixture. Finally, 0.28 v/v% tetramethylethylenediamine (Sigma Aldrich, USA, catalog number T22500) was added to start the polymerization process. After hardening of the PAG, the water-filled balloons were drained and removed [Fig. 3(c)]. The resulting wall-less channels with an 8-mm diameter can be filled with liquids with varying sound speeds. This allows one to not only investigate the SOS spatial resolution and accuracy but also the SOS contrast resolution and the effect of large impedance mismatches on the US reconstructions. Figures 3(d)–3(g) show pictures of the object and MRI cross sections of the object acquired with a Siemens Magnetom Aera 1.5T scanner.

3.1.5.

Channel object

A test object consisting of a breast tissue-mimicking material with four embedded channels was developed to investigate the PA imaging intensity and the QPA performance as a function of depth. All channels have another radius of curvature and are placed under 45 deg separation angles with respect to each other in the $xy$ plane to minimize shadowing effects from the shallower channel(s) on the deeper-lying channel(s). The distance from the channels to the object’s surface gradually increases, and the deeper-lying channel starts at a depth where the more shallow one ends. This together with the rotational symmetry of the PA system enables the extraction of continuous information about the depth dependence of the IQC from 5 to 60 mm. Connection of the channels to a flow system equipped with a blood oxygenator allows flushing the object with blood and to regulate the blood oxygenation levels as in Refs. 41 to 43.

To give the channels the designed curvature, a plate with ridges in the designed channel shapes was 3D printed from PLA. Metal wires (1 mm) were bent along these ridges. US probe sleeves with a 5-mm diameter (Microtek Medical B.V, The Netherlands, catalogue number G88500) were spanned over the metal wires and will later form the walls of the channels and prevent diffusion of blood or other liquids into the test object [see Fig. 4(a)]. The same nylon breast mould introduced for the PA spatial resolution object was also used to shape this test object. The channel shapes were positioned in this mould using a holder [see Fig. 4(b)], and the mould was topped off with a gel wax mixture, which was prepared according to the methodology also used by^58–60 using 93.5 w/w% native gel wax (Mindset, UK), 6 w/w% with paraffin wax (Sigma Aldrich, USA, catalog number 327204), 0.5 w/w% glass beads ($\le 106\mu \mathrm{m}$, Sigma-Aldrich, USA), and 0.5 w/v% ${\mathrm{TiO}}_2$ (Sigma-Aldrich, USA, Catalog No. 248576). The glass beads and ${\mathrm{TiO}}_2$ were added for acoustic and optical scattering, respectively. After solidification of the gel wax, the metal wires were gently removed while leaving the sleeves in place [see Fig. 4(c)]. Finally, the entire object was removed from the mould and the object was placed in a size 8 breast supporting cup to install it in the imaging aperture. The sound speed and acoustic attenuation of the gel wax mixture were measured at 1 MHz and yielded $1445\pm 1\mathrm{m}/\mathrm{s}$ and $0.35\pm 0.1\mathrm{dB}/\mathrm{cm}$, respectively. Optical absorption and scattering were also measured. The optical absorption yielded 0.50 and $0.81{\mathrm{cm}}^{-1}$ at 755 and 1064 nm, and reduced scattering coefficients of 14.76 and $9.78{\mathrm{cm}}^{-1}$ were found. See also Fig. S4 in the Supplementary Material for how the measured properties relate to properties that can be expected in a real breast.

3.2.1.

Black cups

Black cups with sizes 2, 5, and 8 were placed in the PAM 2 system and were imaged with 1064 nm. The cups were filled with water to assure acoustic coupling on both the top and underside of the cup. Images were reconstructed using a 1-SOS model ($1496\mathrm{m}/\mathrm{s}$). Figure 5 shows the reconstructions of the measurements as maximum intensity projections (MIPs) along the $z$ axis. The three projections are normalized to the same maximum value to illustrate absolute amplitude differences on the surface of the cups. With increasing cup size, lower amplitudes are reconstructed since the light is distributed over a larger surface area. For all cups, a low-intensity circle is observed in the center, which may be caused by a lack of detectors around the rotational axis of the PAM 2 system as this space is occupied by the bottom illumination beam. The side illumination beams create a ring-shaped profile on the cups, which can be observed on cup eight, whereas the bottom beam illuminates the cups in the central region. Smaller cups achieve a more homogeneous illumination profile since larger distances between the laser fibers and the cup surface allow the laser beams to diverge more. For the smaller cups, this means that the two illumination regions overlap (see cup 2), whereas they are separated for the larger cups (see cup 8). Homogeneous distributions over the entire cup surface will have the advantage that PA images of the breast volume are obtained with a relatively large field of view and with PA amplitudes that decrease with depth as can be expected from the tissue optical and acoustic properties. Inhomogeneous light distributions in contrary can result in PA images with limited field of views or with PA amplitudes that not only scale with the depth from the skin but also with the spatially varying intensities on the skin surface.

Fig. 5

(a) MIPs along the $z$ direction of the PA images from measurements on black cup 2, 5, and 8. The projections of the cups are normalized to the same value to illustrate absolute fluence differences between the cup sizes. (b) A schematic of the PAM2 illumination geometry showing the contours of black cups 2, 5, and 8.

3.2.2.

PA spatial resolution object

The PA spatial resolution object was imaged with 1064 nm, and an image was reconstructed using a 2-SOS model where $1496\mathrm{m}/\mathrm{s}$ was assigned to the water and $1413\mathrm{m}/\mathrm{s}$ to the volume within the cup, matching to the SOS of the PVCP. To demonstrate the effect of errors in sound speeds assigned to the grid in the PA reconstruction on the PA resolution, a second reconstruction was made where $1496\mathrm{m}/\mathrm{s}$ was assigned to the entire imaging volume. The PA spatial resolution was calculated for both reconstructions for 84 PA targets. These 84 targets were the targets with no other targets in their close surroundings to allow accurate fitting of the Gaussians.

Figure 6 shows the MIPs along the $y$ axis for the 1-SOS and 2-SOS reconstructions. Both MIPs are normalized to their own maximum value. Figure 6 also presents the FWHM of the Gaussians fitted to the PSFs of the targets in $x$, $y$, and $z$ as a function of depth. For both the 1-SOS and 2-SOS reconstruction, the deep-lying targets are reconstructed with higher amplitudes than the more superficially located targets. This is likely a result of the overlapping laser beams and reception fields of the detectors in the center of the imaging volume. The overlapping reception fields of the detectors can also explain why a decreasing FWHM with depth in the 2-SOS reconstruction is observed. For the 1-SOS reconstruction, an increasing trend in the FWHM was found since the reconstruction errors as a result of incorrect sound speeds accumulate with depth.

Fig. 6

(a), (b) MIPs along the $y$ axis for 1-SOS and 2-SOS reconstructions. Both MIPs are normalized to their own maximum pixel value. (c), (d) The FWHM from the Gaussians fitted to the reconstructed beads in the $x$, $y$, and $z$ planes for both the 1-SOS and 2-SOS reconstructions.

3.2.3.

Channel object

Two separate measurements on the channel object were performed to demonstrate its use in assessing the imaging depth. One measurement used 755 nm excitation and the other 1064 nm. During the measurements, the four channels of the object were filled India ink solutions with blood simulating optical absorption coefficients of 2.85 and $4.56{\mathrm{cm}}^{-1}$ for the 755- and 1064-nm measurement, respectively.⁶² Reconstructions were made with a 2-SOS model where a 1445 m/s sound speed was assigned to the test object volume and resulted in the best reconstructions assessed by eye. The SOS assigned to the water yielded $1497\mathrm{m}/\mathrm{s}$.

MIPs along the $y$ and $x$ axis of 2.4-cm thick subvolumes of the channel object centralized at $y=0$ and $x=0$, respectively, are presented in Fig. 7. All projections are normalized to the same value to compare the PA amplitude between the measurements. The most superficial channel was positioned in line with the $xz$ plane and therefore shows in the MIP along the $y$ axis, whereas the deeper-lying channel shows in the MIP along the $x$ axis. Evident signatures from the two deepest channels were not observed in both reconstructions. The points deep inside the object (indicated by the red arrow in the 1064-nm measurement) presumably originate from parasitic PA signals.

Fig. 7

MIPs of the reconstructed channel phantom along the $y$ and the $x$ axes for measurements with two excitation wavelengths showing the two most superficially located channels. All reconstructions are normalized to the same value. The reconstructions are masked by the cup contour and the area in between the two dotted lines is the area where the channels gradually increase from the phantom’s surface. The distance of the channels to the object surface were measured and are shown in the reconstructions.

The shortest distance to the surface was calculated for the channels at the start and end locations of the area where the distance to the surface gradually increases (area in between dashed lines in Fig. 7). The most superficial channel was found to start at 7 and ends at 22 mm and the second channel was found to start at 17 and ends at 42 mm. The most superficial channel is reconstructed as two curves, which is likely caused by a combination of limited optical penetration depths ($\pm 1\mathrm{mm}$) in the ink solutions and by limited bandwidths of the detectors.

In line with the optical properties of the gel wax and the ink solutions, higher PA amplitudes are achieved at 1064 nm. The two largest channels can be observed entirely by eye for this wavelength, corresponding to an imaging depth of 42 mm. In the 755-nm measurement, the most superficial channel can also be observed entirely while the channel in the $zy$ plane only shows up to a depth of 26 mm. The inhomogeneous illumination profile provided by the PAM 2 system on cup 8 can be observed in the reconstructions as higher PA amplitudes are observed around the rotational axis and at the upper sides of the channel. The amplitudes in the region in between those is lower, similarly as was seen on the reconstruction of black cup 8. This demonstrates that the imaging depth of multispectral systems and systems with an inhomogeneous illumination profile will be not only be wavelength-dependent but also spatially variant, which can make a generalized single value for the imaging depth unrepresentative for the entire imaging volume.