2.1.

System Description

The LOIS clinical setup is shown in Fig. 1, and the corresponding schematic is shown in Fig. 2 . The Q-switched Alexandrite laser (Ta2, Light Age, Inc., Somerset, New Jersey) was used as an NIR light source. The laser emitted $75\text{-}\mathrm{ns}$ pulses ( $750\mathrm{mJ}$ /pulse) at $757\mathrm{nm}$ with the repetition rate of $10\mathrm{Hz}$ . The output laser beam was coupled into a custom-made optical fiber bundle and expanded with the lens system, producing a Gaussian profile on the surface of the breast. The laser beam incident on the surface of the breast was $70\mathrm{mm}$ in diameter and had a maximum fluence of $10\mathrm{mJ}∕{\mathrm{cm}}^2$ .

Fig. 1

LOIS-64 in a clinical setup.

Fig. 2

Schematic diagram of LOIS-64.

The acoustic detector probe (Fig. 3 ) represented a hemicylindrical cup in which the breast was suspended through the circular opening on the surface of the patient examination table. The probe cup had the radius of its cylindrical surface of $70\mathrm{mm}$ and width of $90\mathrm{mm}$ , allowing accommodation of a large breast. The arc-shaped imaging array consisted of 64 rectangular polyvinylidene fluoride (PVDF a polymer piezoelectric material; Measurement Specialties, Inc., Hampton, Virginia) transducers $(20\times 3\times 0.11\mathrm{mm})$ and was used for reconstruction of 2-D images (mediolateral or craniocaudal), representing changes in optical absorption of the breast tissue within the imaging slice passing through the imaging array. The illumination with a wide laser beam orthogonally to the imaging array allowed small changes of optical fluence across the area of interest within the imaging slice, which was essential for good quality of the OA image.

Fig. 3

Optoacoustic probe for LOIS-64.

The electrical signals generated by each transducer were amplified using a two-stage amplifier designed in house to respond to the wide bandwidth of the OA signals. The first stage, a charge amplifier, was designed for a low-noise amplification of microvolt signals generated by a capacitive source (OA transducer). The gain of this stage was $30\mathrm{dB}$ . Two of these preamplifier boards were used for the imaging array and were embedded in the acrylic body of the probe. The probe was covered with an electromagnetic (EM) shield to protect OA signals from any transmitted EM noise. The second stage, a standard $30\text{-}\mathrm{dB}$ signal amplifier, was followed by an analog-to-digital converter (ADC) board based on $12\text{-}\text{bit}$ ADCs (AD9042, Analog Devices, Norwood, Massachusetts), which digitized the OA signals with a maximum sampling rate of $41\mathrm{MHz}$ . The data acquisition system communicated with a PC through the Ethernet interface and was controlled by a field-programmable gate array (Stratix II, Altera, San Jose, California). The LOIS-XP software allowed programmable control of the data acquisition (sampling rate and size of the data packets), system configuration (probe shape and size, the number of transducers, signal and image processing), and the real-time display of the signals from the individual transducers or the reconstructed images. The frame rate in the real-time imaging mode could be up to $1\text{frame}\text{per}\text{second}$ for $512\times 512\text{pixel}$ images and $10\text{frames}\text{per}\text{second}$ for $128\times 128\text{pixel}$ images.

2.2.

Measuring Impulse Response of an Acoustic Transducer

Impulse response of a single acoustic transducer was measured as an electrical response to a planar acoustic wave generated by the $6\text{-}\mathrm{ns}$ laser pulse absorbed in a layer of india ink (Sanford Higgins Fountain Pen india Ink, Sanford, Bellwood, Illinois). The laser light was emitted at the wavelength of $1064\mathrm{nm}$ by the Nd:YAG laser (Brilliant-B, Quantel, Bozeman, Montana). The concentrated india ink had an absorption coefficient ${\mu }_a\approx 1800{\mathrm{cm}}^{-1}$ at $1064\mathrm{nm}$ (extrapolated from the measurements on the diluted solution using a spectrophotometer—DU 520, Beckman Coulter, Fullerton, California). The schematic of the test setup is shown in the Fig. 4a . The acrylic block (transparent for IR light) had an acoustic impedance about two times that of the india ink,²⁹ and it was used to create a hard acoustic boundary at the light-absorbing interface. The resultant short, monopolar, biexponential acoustic signal had essentially flat amplitude spectrum within the frequency band of interest [ $0\text{to}10\mathrm{MHz}$ ; Fig. 4b], therefore representing the input $\delta$ -function. The acoustic signal was sensed by the PVDF transducer and recorded using an oscilloscope (TDS 3014, Tektronix, Beaverton, Oregon) with 10,000 samples at the sampling rate of $1\mathrm{GHz}$ .

Fig. 4

(a) Setup for measurements of the transducer impulse response. (b) Amplitude spectrum of the acoustic signal generated during measurements of the transducer impulse response. Computer simulation.

2.3.

Measuring Sensitivity of a LOIS Channel

The calibrated point acoustic emitter (NP/OA90-3, Dapco Industries, Ridgefield, Connecticut) was positioned in the imaging plane of the water-filled acoustic detector probe at a distance of $60\mathrm{mm}$ from the transducer array. A pivotal mechanism with a point of rotation located in the center of the probe was used to position the emitter over a specific transducer of the probe. The emitter was driven with a $20\text{-}\text{cycle}$ $1.5\text{-}\mathrm{MHz}$ harmonic burst. The peak-to-peak driving voltage of $970\mathrm{mV}$ provided pressure amplitude of $85\mathrm{Pa}$ at the surface of the transducer.

2.4.

Directivity of an Acoustic Transducer

The transducer sensitivity also depends on the incident angle of the acoustic wave and is described by the directivity of the transducer. For a finite-size transducer, the measured acoustic signal is a temporal convolution of the pressure signal at the transducer surface $p^{\prime }\left(t\right)$ over the temporal transducer window function $T(\theta ,t)$ :

The signal received by a transducer at the right angle is not distorted and retains its amplitude. Therefore, the directivity function for a transducer $D_{\mathrm{tr}}\left(\theta \right)$ was defined as the ratio of the signal maximum at the incident angle $\theta$ to the signal maximum at $\theta =90\mathrm{deg}$ . Using the temporal transducer window function 2, the directivity function for an incident N-shaped acoustic pulse generated by a spherical source^{21, 22} can be calculated as:

2.5.

Sensitivity of the Whole LOIS-64 System

The sensitivity of the LOIS-64 system characterizes its capability to visualize small, weakly absorbing objects located in the imaging slice of the array of transducers.

As a simplified example, let us consider a spherical tumor. For this model tumor, the maximum magnitude of the temporal pressure integral (TPI) signal $\left(U_{\max }\right)$ measured by a single transducer can be estimated as:

We graphically visualized the total sensitivity of the LOIS-64 using the spatial distribution of the minimal tumor diameter $d_0^{\min }\left(\vec{r}\right)$ detectable inside the imaging slice:

We assumed that a safe (clinically approved by ANSI) incident laser fluence $(E_0=20\mathrm{mJ}∕{\mathrm{cm}}^2)$ is used during the optoacoustic imaging. We estimated the value of the Gruneisen coefficient for a breast tumor $\Gamma \approx 0.24$ using the following tissue material properties:³² volumetric thermal expansivity ${\alpha }_p\approx 4\cdot {10}^{-4}{\mathrm{K}}^{-1}$ ; speed of sound $c_s\approx 1500\mathrm{m}∕\mathrm{s}$ ; specific heat capacity $c_p\approx 3.7\cdot {10}^3\mathrm{J}∕(\mathrm{kg}\cdot \mathrm{K})$ . We estimated ${\mu }_{\mathrm{eff}}\approx 1.1{\mathrm{cm}}^{-1}$ using the following optical properties of the breast tissue:¹⁷ ${\mu }_s^{\prime }=9.5{\mathrm{cm}}^{-1}$ and ${\mu }_a=0.04{\mathrm{cm}}^{-1}$ . We used the differential optical absorption coefficient for breast tumors relative to surrounding normal breast tissue $\Delta {\mu }_a\approx 0.05{\mathrm{cm}}^{-1}$ .^{17, 18}

2.6.

Noise of the LOIS-64 System and the Signal Processing

There are two types of noise, that affect the quality of the OA images obtained with LOIS: the first type is the electric noise, as represented by the external EM sources, and thermal fluctuations of the transducers and amplifiers.²⁷ The electric noise is reduced in LOIS by proper EM shielding and by limiting the high-frequency part of the amplifier bandwidth. The second type of noise comes from the OA sources, acoustic reflections, and probe resonance. Noise-generating OA sources are created by the optical energy absorbed within the skin and the first $\sim 10\mathrm{mm}$ of breast tissue. They dramatically expand the dynamic range of the optoacoustic signals, thereby reducing the resolution and contrast of the image. Reflection of ultrasonic waves generated by strong OA sources from the acoustic boundaries, including echogenic areas inside the breast, physical breast boundaries, and the housing of piezoelectric transducers, also can be a source of acoustic noise. Last, an acoustically excited probe can resonate and create large low-frequency (LF) artifacts.

Acoustic reflections and signals generated by the probe were minimized in LOIS-64 by utilizing the probe material, which had proper acoustic matching to the breast tissue, and large acoustic attenuation. Narrowing the transducer directivity orthogonally to the imaging plane (see Sec. 3.2 helped to reduce OA signals generated on the illuminated surface of the breast and by other OA sources located outside the imaging plane.³³

An important feature of the optoacoustic signal used for the designing of an optimal signal processing was the bipolar shape (N-shape) of the measured pressure pulses with a very short rise time and, simultaneously, significant energy residing in the low-frequency band (below $1\mathrm{MHz}$ ).²⁷ Additionally, the visualized breast tumors could be anywhere between $5\mathrm{mm}$ and $20\mathrm{mm}$ in size, generating N-shaped pressure pulses with different spectral bands. Therefore, in order to eliminate-LF acoustic artifacts and simultaneously transform the bipolar pressure pulse to the monopolar pulse suitable for the tomographic reconstruction of the OA image, we implemented the wavelet transform using a wavelet family resembling the N-shaped OA signal.³⁴ The wavelet transform has been established in signal processing as a superior tool for pattern recognition and temporal localization of specified signal patterns.³⁵ We found that the third derivative of the Gaussian wavelet [Fig. 5a ] was the best candidate for filtering the N-shaped signals [Fig. 5b], providing a monopolar pulse with enhanced sharpness of the boundaries [Fig. 5c], while significantly reducing LF acoustic artifacts [Fig. 13d].³⁶ In the frequency domain, the chosen wavelet had a narrow bandpass region and a steep slope in the LF band, which allowed more precise recovery of the OA signals by using a large number of wavelet scales (up to nine scales, with filter lengths ranging from 4 to 1024 data samples).

Fig. 5

(a) Third derivative of the Gaussian wavelet. (b) N-shaped OA pressure signal. (c) OA pressure signal transformed with nine scales of the wavelet.

Fig. 13

Processing of the OA signals from the tumor phantom. (a) Original signal showing the small N-shaped pulse riding on the large LF acoustic artifact. (b) Integration of the measured pressure wave form showing amplification of the LF acoustic artifacts. (c) Application of the $75\text{-}\mathrm{kHz}$ FIR high-pass filter with Hamming window followed by the integration. (d) Eight-scaled transform using the third derivative of the Gaussian wavelet; retains the short rise time of the tumor signal. (e) Six-scaled transform using the third derivative of the Gaussian wavelet; highlights the edges of the tumor signal. (f) Differentiation of the measured pressure wave form; reduces signal-to-noise ratio.

2.7.

Reconstruction of OA Images

To reconstruct a 2-D OA image from pressure signals measured by the imaging array, we used the radial back-projection algorithm.^{11, 24} Briefly, the algorithm employs temporal integration of the pressure signals. Then each sample $s$ of the digitized TPI is added to the corresponding pixels of the image located on the intersection of the imaging plane with spherical surface of radius $R=c_s\Delta t(s-1)$ , where $\Delta t$ is the sampling period. Additionally the algorithm weighs the values according to the directivity diagram of each transducer (see Sec. 2.4).

2.8.

Computer Simulation of the Optical Energy Absorbed by the Breast Tissue

Three-dimensional distribution of the optical properties (index of refraction, absorption coefficient, scattering coefficient, and scattering anisotropy) simulating breast tissues inside the acoustic detector probe was the core of the model based on the Monte Carlo method.³⁷ The modeling was used to generate 3-D distribution of the optical energy absorbed within the breast tissue. Table 1 contains the values of the optical parameters used, and Fig. 6 explains the structural features of the model. The modeling was performed using 10 million photons for the $70\text{-}\mathrm{mm}$ -diam Gaussian laser beam. The beam was incident orthogonally to the skin at $x=0\mathrm{mm}$ and $y=-35\mathrm{mm}$ from the center of the probe.

Fig. 6

Schematic of the model used to simulate the distribution of the optical energy absorbed by the breast tissue during optoacoustic imaging with LOIS-64. Skin layer was $1\mathrm{mm}$ thick. Tumor was $5\mathrm{mm}$ in diameter and located $z=20\mathrm{mm}$ from the skin; $x=-10\mathrm{mm}$ and $y=-30\mathrm{mm}$ from the center of the OA probe

Table 1

Optical properties of the breast used in modeling of absorbed optical energy (Refs. 17, 31).

2.9.

Phantoms Used in Studies of the OA Image Contrast and Resolution

We used a phantom based on the aqueous milk solution representing a female breast with a tumor to study the acoustic artifacts and the contrast of OA images. The aqueous milk solution was chosen because it had acoustic properties close to those of the breast tissue and was easy to prepare with desired optical properties. The required optical absorption coefficient was achieved by the dissolved india ink. The optical properties of the liquid phantoms used in experiments were measured with an integrating sphere (RT-6, SphereOptics, Concord, New Hampshire). We filled the LOIS-64 probe with solution having optical absorption coefficient ${\mu }_a=0.04{\mathrm{cm}}^{-1}$ and reduced scattering coefficient ${\mu }_s^{\prime }=9{\mathrm{cm}}^{-1}$ . For studies of the image contrast and OA artifacts, we positioned a thin-wall rubber shell ( $10\mathrm{mm}$ in diameter, $10\text{-}\mu \mathrm{m}$ wall) filled with the solution mimicking the tumor ( ${\mu }_a=0.15{\mathrm{cm}}^{-1}$ , ${\mu }_s^{\prime }=12{\mathrm{cm}}^{-1}$ ) inside the imaging slice of the probe about $20\mathrm{mm}$ below the illuminated surface (Fig. 7 ). For studies of the image resolution, we positioned two parallel plastic tubes (inside diameter $0.7\mathrm{mm}$ , wall thickness $0.05\mathrm{mm}$ ; Zeus, Orangeburg, South Carolina) filled with the india ink dissolved in an aqueous milk solution ( ${\mu }_a=2.8{\mathrm{cm}}^{-1}$ , ${\mu }_s^{\prime }=4{\mathrm{cm}}^{-1}$ ) inside the imaging slice of the probe [Fig. 8a ]. The ink-filled interiors of the tubes were separated by $0.5\mathrm{mm}$ [Fig. 8b]. The incident laser fluence was $1.4\mathrm{mJ}∕{\mathrm{cm}}^2$ .

Fig. 7

Experimental setup for studies of the OA image contrast.

Fig. 8

(a) Experimental setup for studies of the OA image resolution. (b) Two parallel plastic tubes filled with the india ink solution separated by $0.5\mathrm{mm}$ used to simulate blood vessels.

2.J.

Preliminary Clinical Studies on Breast Cancer Patients

Preliminary clinical studies of the LOIS-64 were held in the Radiology Department at the University of Texas Medical Branch (Galveston, Texas) under institutional review board (IRB) Protocol #99464. A total of 42 patients participated in the studies. In the first phase, the safety studies were performed on 15 patients scheduled for radical mastectomy. The pathology studies performed on breasts surgically excised after LOIS imaging demonstrated safety of laser illumination and absence of thermomechanical damage to tissues.¹⁰ In the second phase, a patient with a suspicious breast region identified on the mammography and/or ultrasonic image was scheduled for breast biopsy. Prior to the biopsy, an imaging with LOIS was performed. The recorded OA signals were processed with the 8-scale wavelet transform and used in the reconstruction of the OA image according to the procedure described in the Sec. 2.7. The linear grayscale palette of the resultant OA image was individually optimized in each case to get the maximum image contrast. The OA images were visually analyzed and compared to the mammography and ultrasonic images. We considered that the tumor can be seen on the OA image if an isolated area of increased intensity could be localized in the quadrant of the breast suspected to contain a tumor according to the ultrasound and mammography images. The final diagnosis was made based on the biopsy results.

During the second phase, the LOIS prototype hardware, software, and imaging procedure were continuously optimized for the best performance, with the latest design presented in this work. The details on the technical parameters and performance of some intermediate versions of LOIS can be found in our previous publications.^{10, 11, 23, 38} Due to system modifications, no statistical inferences could be made during the second phase of the preliminary clinical studies. In this work, we report the capability of LOIS to visualize breast cancer based on the 27 cases studied during the second phase of the preliminary clinical studies. Two cases that demonstrate the feasibility of clinical application of the LOIS-64 are discussed in greater detail.

3.1.

LOIS Frequency Response

Using the technique described in Sec. 2.2, we measured the impulse response of an individual acoustic transducer [Fig. 9a ]. Acoustic backing of PVDF transducers was provided by an acoustically matching scattering and absorbing composite, which caused no reverberations after detection of a pressure pulse [Fig. 9a]. The frequency response of PVDF transducers is shown in the Fig. 9b, having a bandwidth up to $2.5\mathrm{MHz}$ . Because both amplification stages were operated in the wideband mode, the system frequency response followed that of the PVDF transducer.

Fig. 9

(a) Impulse response of a PVDF transducer used in LOIS-64. (b) The corresponding amplitude response $(f_{\max }\approx 2.5\mathrm{MHz})$ .

3.2.

Sensitivity of an Individual LOIS Channel and the Whole LOIS-64 System

Using the methodology described in Sec. 2.5, we measured the average sensitivity of an individual LOIS channel to a normally incident $1.5\text{-}\mathrm{MHz}$ acoustic wave to be $1.66\mathrm{mV}∕\mathrm{Pa}$ , with a standard deviation of $0.21\mathrm{mV}∕\mathrm{Pa}$ $(N=46)$ . The sensitivity of an individual LOIS channel to normally incident acoustic waves at other frequencies could be derived from the transducer frequency response shown in the Fig. 9b.

For the N-shaped pressure wave incident parallel to the long side of the planar rectangular transducer, the convolution (1) leads to a narrow directivity [ $\pm 15\mathrm{deg}$ , based on the $-6\text{-}\mathrm{dB}$ level; see Fig. 10a ]. Our experimental data (crosses) confirmed that the LOIS-64 produces 2-D images of slices with a thickness of about the length of the long side of the transducers $\left(20\mathrm{mm}\right)$ . On the other hand, the short side of LOIS transducers $\left(3\mathrm{mm}\right)$ provides directivity close to $\pm 60\mathrm{deg}$ within the imaging slice [Fig. 10b].

Fig. 10

Directivity of individual transducers used in LOIS-64. OA target is a $9.5\text{-}\mathrm{mm}$ polyvinyl chloride–plastisol ball. Graphs made in the polar coordinate system show the amplitude of the measured OA signal as a function of the incident angle (deg). The data is normalized to the amplitude of the OA signal measured for orthogonally incident acoustic wave. (a) Directivity in the plane parallel to the long side of the transducer $\left(20\mathrm{mm}\right)$ ; (b) directivity in the imaging plane.

The measured RMS noise of the TPI signal was $0.22\cdot {10}^{-7}\mathrm{V}\cdot \mathrm{s}$ . Using Eq. 7, we calculated that for a single-pulse data acquisition. a tumor with diameter of $6\text{to}11\mathrm{mm}$ could be detected by LOIS-64 through the entire imaging slice (Fig. 11 ).

Fig. 11

Minimal diameter of a tumor detectable by LOIS-64 in the imaging slice. Signal-to-noise ratio is 1; single-pulse acquisition. Computer simulation. Minimal diameters of detectable tumors are linearly scaled on the image palette from $6\mathrm{mm}\text{to}11\mathrm{mm}$ . Area within the imaging slice painted with a particular color specifies possible locations of the center of the tumor, which would be detected by the array, providing its diameter is greater than or equal to the one coded by that color. (Color online only.)

3.3.

OA Artifacts, Signal Processing, and Image Contrast

Figure 12 displays the Monte Carlo simulation of 3-D distribution of the optical energy absorbed by the breast within the LOIS-64 probe, illustrating the nature of LF acoustic artifacts. The large volume in the front part of the breast where laser light enters the tissue contains a significant amount of the absorbed optical energy. Also part of the probe’s surface is receiving some light, which creates strong OA signals. Both of those phenomena resulted in large-magnitude LF OA signals that dominated the much smaller N-shaped pulse generated by the $5\text{-}\mathrm{mm}$ spherical tumor located in the imaging slice just $20\mathrm{mm}$ below the skin [Fig. 13a ]. The corresponding TPI signals [Fig. 13b], used for the image reconstruction, provided low-pass filtering, therefore enhancing the LF artifacts and resulting in a very poor contrast of the reconstructed OA image [Fig. 14a ]. Figure 13c displays the signal, and Fig. 14b displays the image obtained by filtering with the optimized $75\text{-}\mathrm{kHz}$ finite impulse response (FIR) high-pass filter with the Hamming window followed by the integration. Despite the removed LF artifacts, the subsequent integration significantly reduced the rise time of the signal and the sharpness of the OA image. Application of the 8-scaled wavelet transform made the tumor signal clearly visible [Fig. 13d], and significantly improved the tumor contrast on the reconstructed OA image [Fig. 14c]. Using fewer scales in the wavelet transform highlighted fast-changing signals from the boundaries of the tumor and flattened the background even more [Figs. 13e and 14d]. However, the contrast of the tumor interior disappeared [Fig. 14d]. A very low signal-to-noise ratio(SNR) can be observed when a post-processing differentiation is used in the attempt to enhance high-frequency features of the OA signals [Fig. 13f], as is done in some algorithms used in the OAT.³⁹ The differentiation results in a poor contrast of the image but enhances boundaries of the object in a way similar to the low-scaled wavelet transform [Fig. 14e]. Also, in this case, the image contrast could be improved if differentiation of the pressure signals is done by the hardware. The effect of averaging on the tumor contrast was demonstrated in Fig. 14c (256 averaged acquisitions) versus Fig. 14f (no averaging). The arc-shaped artifacts seen around the tumor phantom are caused by the powerful OA signal generated at the air–fluid interface and reflected from the phantom target toward the transducers. When reconstructed using a regular radial back-projection technique, such signals will produce a distorted image of the target. The artifacts still persist after the applied signal processing [Figs. 14b, 14c, 14d, 14e, 14f].

Fig. 12

Monte Carlo simulation of the optical energy absorbed by the breast within the LOIS-64 probe. (a) Front view, showing Gaussian transverse profile of the laser energy absorbed within the breast and the energy absorbed by the surface of the acoustic probe. (b) Side view, showing distribution of the laser energy absorbed by the breast tissues along the direction of light propagation as well as the tumor.

Fig. 14

OA images of the tumor phantom. (a) RBP with the integration of the measured pressure wave form. (b) Application of the $75\text{-}\mathrm{kHz}$ FIR high-pass filter with Hamming window followed by the integration, showing high tumor contrast but significant blurring. (c) 8-scale transform using the third derivative of the Gaussian wavelet; retains the high contrast and minimizes blurring of the tumor. (d) Six-scaled transform using the third derivative of the Gaussian wavelet; highlights the tumor boundaries. (e) RBP using the differentiated pressure wave form; provides sharp images but with low contrast. (a) to (e) Acquisition with 256 averaged laser pulses. (f) Eight-scaled transform using the third derivative of the Gaussian wavelet. Acquisition without averaging.

3.4.

Resolution of the OA Images

Figure 15 displays OA images obtained with the resolution phantom using high-pass filtering and wavelet-based signal processing. Using the optimized high-pass filter with a cutoff frequency around $400\mathrm{kHz}$ allowed resolution of $0.5\mathrm{mm}$ with horizontal orientation of the tubes [Fig. 15a]. However, the image contrast was very poor in comparison to the results obtained with the 6-scaled wavelet transform [Fig. 15b]. The high-pass filtering of OA signals followed by integration was unable to resolve $0.5\mathrm{mm}$ with a vertical orientation of the tubes [Fig. 15c], while the 5-scaled wavelet provided clear high-contrast images of the separate tubes [Fig. 15d]. The arc-shaped artifacts, created by the acoustic waves reflected from the resolution phantom, and described in the previous section, are particularly apparent on the images with vertical orientation of the tubes [Figs. 15c and 15d].

Fig. 15

OA images of the resolution phantom. Application of the $400\text{-}\mathrm{kHz}$ FIR high-pass filter with Hamming window optimized for high-resolution imaging followed by the integration; provides $0.5\text{-}\mathrm{mm}$ horizontal resolution (a) but very poor contrast and worse than $0.5\text{-}\mathrm{mm}$ vertical resolution (c). Application of the 6-scaled (b) and 5-scaled (d) transform using the third derivative of the Gaussian wavelet; provides $0.5\text{-}\mathrm{mm}$ horizontal and vertical resolution and high image contrast.

3.5.

Demonstration of the Clinical Performance of the LOIS-64

In the clinical study on 27 patients using LOIS-64, biopsy performed after LOIS procedure revealed 34 lesions, including 26 malignant tumors (25 carcinomas and 1 papilloma with focal area of rapid abnormal cell growth) and 8 benign formations (3 fibroadenomas, 3 fibrocystic lesions, and 2 regions of fibrosis with calcifications). In one dense breast suspected to contain a tumor based on the x-ray mammography, the ultrasound did not find any abnormalities, and biopsy was not performed. The OA images identified (see Sec. 2J for details on the visualization criterion) 22 lesions, including 17 carcinomas, 1 papilloma with focal area of rapid abnormal cell growth, 2 fibroadenomas, 1 fibrocystic lesion, and 1 region of fibrosis with calcifications. Of the eight cases when OA imaging failed to identify the cancer, in two cases, the OA image was reconstructed but the tumor could not be visualized due to insufficient contrast, five cases were due to the system malfunction, and one case was due to operator error. Table 2 summarizes the efficiency of a breast tumor visualization using ultrasound, mammography, and OA imaging. The studied malignant tumors were between 11 and $26\mathrm{mm}$ in size, with the centers located $8\text{to}26\mathrm{mm}$ inside the breast.

Table 2

Summary of the preliminary clinical studies using LOIS.

An example of how LOIS-64 can improve x-ray mammography is depicted in Fig. 16 . The tumor (poorly differentiated infiltrating ductal carcinoma grade $3∕3$ , according to the biopsy) could barely be localized on the mammography image of the radiologically dense breast [Fig. 16a]. The auxiliary ultrasonic imaging helped to identify a $23\times 15\mathrm{mm}$ tumor located at the depth of $21\mathrm{mm}$ [Fig. 16b]; however, the tumor margins were not clear. The tumor was clearly seen on the OA image [Fig. 16c] at the same location as determined by the ultrasound and mammography. The tumor contrast on the OA image exceeded the tumor contrast on the ultrasonic image (judged by the visual comparison of the images created using the same linear grayscale palette) and was attributed to the associated microvasculature.

Fig. 16

Example of the clinical images showing the breast cancer: (a) mediolateral mammography, (b) ultrasonic, and (c) mediolateral optoacoustic images. High contrast of the object in the OA image implies the advanced angiogenesis indicative of a malignant tumor.

In another case, the mammography image in Fig. 17a revealed the round shape of the tumor (about $15\mathrm{mm}$ in diameter located $16\mathrm{mm}$ under the skin), suggesting that it was benign. However, the LOIS image [Fig. 17c] exposed a focal area of enhanced brightness indicative of the advanced angiogenesis. The additional imaging with the color Doppler ultrasound [Fig. 17b], confirmed enhanced blood flow in one focal area of the tumor. The biopsy of this tumor revealed the intraductal papilloma with some clusters of rapidly growing cells of unusual (abnormal) pathology, indicating possible malignancy.

Fig. 17

(a) Digital mammography image (mediolateral projection) of the breast, showing a round, apparently benign tumor. (b) Ultrasonic imaging with the Doppler showing the tumor boundaries and the focal blood flow increase. (c) Optoacoustic image (mediolateral projection) showing the enhanced contrast around the area of the focal blood flow increase detected by the Doppler ultrasound.