We report a coregistered ultrasound and photoacoustic imaging protocol for the transvaginal imaging of ovarian/adnexal lesions. The protocol may be valuable to other translational photoacoustic imaging studies, especially those using commercial ultrasound arrays for the detection of photoacoustic signals and standard delay-and-sum beamforming algorithms for imaging.
Ovarian cancer remains the deadliest of all the gynecological malignancies due to the lack of reliable screening tools for early detection and diagnosis. Photoacoustic imaging or tomography (PAT) is an emerging imaging modality that can provide the total hemoglobin concentration (relative scale, rHbT) and blood oxygen saturation (%sO2) of ovarian/adnexal lesions, which are important parameters for cancer diagnosis. Combined with coregistered ultrasound (US), PAT has demonstrated great potential for detecting ovarian cancers and for accurately diagnosing ovarian lesions for effective risk assessment and the reduction of unnecessary surgeries of benign lesions. However, PAT imaging protocols in clinical applications, to our knowledge, largely vary among different studies. Here, we report a transvaginal ovarian cancer imaging protocol that can be beneficial to other clinical studies, especially those using commercial ultrasound arrays for the detection of photoacoustic signals and standard delay-and-sum beamforming algorithms for imaging.
Photoacoustic imaging or tomography (PAT) is a hybrid imaging modality that measures the optical absorption distribution at US resolution and depths far beyond the tissue optical diffusion limit (~1 mm). In PAT, a nanosecond laser pulse is used to excite biological tissue, causing a transient temperature rise due to optical absorption. This leads to an initial pressure rise, and the resultant photoacoustic waves are measured by US transducers. Multispectral PAT involves the use of either a tunable laser or multiple lasers operating at different wavelengths to illuminate the tissue, thereby enabling the reconstruction of optical absorption maps at multiple wavelengths. Based on the differential absorption of oxygenated and deoxygenated hemoglobin in the near-infrared (NIR) window, multispectral PAT can compute the distributions of oxygenated and deoxygenated hemoglobin concentrations, the total hemoglobin concentration, and the blood oxygen saturation, which are all functional biomarkers related to tumor angiogenesis and blood oxygenation consumption or tumor metabolism. PAT has demonstrated success in many oncology applications, such as ovarian cancer1,2, breast cancer3,4,5, skin cancer6, thyroid cancer7,8, cervical cancer9, prostate cancer10,11, and colorectal cancer12.
Ovarian cancer is the deadliest of all gynecological malignancies. Only 38% of ovarian cancers are diagnosed at an early (localized or regional) stage, where the 5 year survival rate is 74.2% to 93.1%. Most are diagnosed at a late stage, for which the 5 year survival rate is 30.8% or less13. Current clinical diagnosis methods, including transvaginal ultrasonography (TUS), Doppler US, serum cancer antigen 125 (CA 125), and human epididymis protein 4 (HE4), are shown to lack sensitivity and specificity for early ovarian cancer diagnosis14,15,16. Additionally, a large portion of benign ovarian lesions may be difficult to diagnose accurately with current imaging technologies, which leads to unnecessary surgeries with increased healthcare costs and surgical complications. Thus, additional accurate non-invasive methods for the risk stratification of adnexal masses are needed to optimize the management and outcomes. Clearly, a technique that is sensitive and specific to early-stage ovarian cancer and more accurate in identifying malignant from benign lesions is needed.
Our group has developed a coregistered transvaginal US and PAT system (USPAT) for ovarian cancer diagnosis by combining a clinical US system, a custom-made probe sheath to house the optical fibers for light delivery, and a tunable laser1. The total hemoglobin concentration (relative scale, rHbT) and the blood oxygen saturation (%sO2) derived from the USPAT system have demonstrated great potential for the detection of early-stage ovarian cancers and for accurately diagnosing ovarian lesions for effective risk assessment and the reduction of unnecessary benign lesion surgeries1,2. The current system schematic is shown in Figure 1, and the control block diagram is shown in Figure 2. This strategy has the potential to be integrated into existing TUS protocols for ovarian cancer diagnosis while providing functional parameters (rHbT, %sO2) to improve the sensitivity and specificity of TUS.
All the research performed was approved by the Washington University Institutional Review Board.
1. System configuration: Optical illumination (Figure 1)
2. System configuration: Ultrasonic detection and scanning scheme
3. System calibration
4. A sample experimental procedure: Transvaginal USPAT imaging of the human ovary
Here, we show examples of malignant and normal ovarian lesions imaged by USPAT. Figure 3 shows a 50 year old premenopausal woman with bilateral multicystic adnexal masses revealed by contrast-enhanced CT. Figure 3A shows the US image of the left adnexa with the ROI marking the suspicious solid nodule inside the cystic lesion. Figure 3B shows the PAT rHbT map superimposed onto the US and shown in red. The rHbT showed extensive diffused vascular distribution in the depth range of 1 cm to 5 cm and the level was high at 17.1 (a.u.). Figure 3C shows the %sO2 distribution superimposed onto the US, and the level was low at a mean value of 46.4%. The histograms of rHbT and %sO2 in the ROI are shown in the right corner of the rHbT and %sO2 maps. Surgical pathology revealed well-differentiated endometrioid adenocarcinoma of both the right and left ovaries.
Figure 4 shows a 46 year old woman with bilateral cystic lesions. Figure 4A shows the US image of the right ovary with a simple cyst measuring 4.2 cm in maximum diameter. Figure 4B shows the PAT rHbT map superimposed onto the coregistered US showing scattering signals on the left side of the lesion with a low average level of 4.8 (a.u). Figure 4C shows the %sO2 map, which revealed a higher %sO2 content of 67.5%. The surgical pathology revealed a normal right ovary with follicular cysts.
Based on the pilot data, malignant ovarian lesions revealed 1.9 times higher rHbT and 9% lower %sO2 on average as compared with benign lesions1. These two representative examples highlight the importance of the functional parameters provided by PAT in the diagnosis of US-detected lesions.
Wavelengths | 750 nm | 780 nm | 800 nm | 830 nm |
Fiber 1 | 4.79 mJ/cm2 | 6.16 mJ/cm2 | 6.59 mJ/cm2 | 6.33 mJ/cm2 |
Fiber 2 | 4.62 mJ/cm2 | 5.39 mJ/cm2 | 5.99 mJ/cm2 | 6.50 mJ/cm2 |
Fiber 3 | 4.79 mJ/cm2 | 6.07 mJ/cm2 | 6.76 mJ/cm2 | 6.84 mJ/cm2 |
Fiber 4 | 4.70 mJ/cm2 | 6.07 mJ/cm2 | 6.67 mJ/cm2 | 6.50 mJ/cm2 |
Total | 18.90 mJ/cm2 | 23.69 mJ/cm2 | 26.01 mJ/cm2 | 26.17 mJ/cm2 |
MPE (ANSI) | 25.2 mJ/cm2 | 28.9 mJ/cm2 | 31.7 mJ/cm2 | 36.4 mJ/cm2 |
Table 1: Representative laser energy density measurements in units of mJ/cm2 coupled to the four fiber tips for four wavelengths along with their corresponding MPE values.
Figure 1: The coregistered US and PAT system and probe. The US system is extended with another monitor for the USPAT display software, and it receives laser triggers to synchronize the US acquisition. The laser beam is expanded by a plano-convex lens (L1), collimated by a plano-concave lens (L2), split into four beams with two stages of beam splitters (BS), and coupled into multi-mode fibers (MMF) with four plano-convex lenses (L3-6) and fiber couplers (FC1-4). The fibers are attached to the endocavity US probe through a custom probe sheath. Mirrors (M) are used to redirect light in the confined space when necessary. The control computer is not shown. Please click here to view a larger version of this figure.
Figure 2: Block diagram of the USPAT control software. The control software automates the imaging process by changing the laser wavelength, sending data acquisition commands to the clinical US system, and signaling the display software to process and visualize the data. The clinical US system receives triggers from the laser directly to synchronize the laser excitation with the US detection. The display software reads the RF data from the file system. Please click here to view a larger version of this figure.
Figure 3: A 50 year old premenopausal woman with bilateral multicystic adnexal masses revealed by contrast-enhanced CT. (A) US image of the left adnexa with the ROI marking the suspicious solid nodule inside the cystic lesion. (B) The PAT rHbT map superimposed onto the US and shown in red. The rHbT showed extensive diffused vascular distribution in the depth range of 1 cm to 5 cm, and the level was high at 17.1 (a.u.). (C) The %sO2 distribution superimposed onto the US. The level was low at a mean value of 46.4%. Surgical pathology revealed well-differentiated endometrioid adenocarcinoma of both the right and left ovaries. The depth was marked on the right side of the B-scan images. Please click here to view a larger version of this figure.
Figure 4: A 46 year old woman with bilateral cystic lesions. (A) US of the right ovary with a simple cyst measuring 4.2 cm in maximum diameter. (B) The PAT rHbT map superimposed onto the coregistered US showing scattering signals on the left side of the lesion with a low average level of 4.8 (a.u). (C) The %sO2 map revealed a higher %sO2 content of 67.5%. The surgical pathology revealed a normal right ovary with follicular cysts. The depth was marked on the right side of the B-scan images. Please click here to view a larger version of this figure.
Supplementary File 1: Probe sheath. Please click here to download this File.
Optical illumination
The number of fibers used is based on two factors: light illumination uniformity and system complexity. It is critical to have a uniform light illumination pattern at the skin surface to avoid hot spots. It is also important to keep the system simple and robust with a minimal number of fibers. The use of four separate fibers has previously been shown to be optimal for creating uniform illumination at depths of several millimeters and beyond. Additionally, the light coupling to four optical fibers is relatively simple and robust, as needed for patient studies. We have previously shown that the use of four 1 mm core multi-mode optical fibers, with the fiber tips approximately 10 mm away from the tissue, housed in a highly reflective probe sheath (refer to Supplementary File 1 for the design) are optimal for transvaginal photoacoustic imaging17.
USPAT display software
The clinical US system we use can be programmed for the real-time display of single-wavelength PAT21. However, our method requires custom post processing of multispectral PAT data to compute functional parameters, so we chose to implement our own USPAT display software in C++ to compute and visualize functional maps and parameters. US and PAT B-mode images are computed from the RF data using standard delay-and-sum beamforming, log compression, and dynamic range and are then interpolated into a fan shape. The rHbT and %sO2 maps computed from the multispectral PAT data (see "Computation of the rHbT and %sO2" later in the discussion) are displayed on the coregistered image or, optionally, in a user-defined region of interest (ROI). The mean and maximum of the %sO2 and rHbT are displayed on the screen for reference. During imaging, the display software is used in server mode to listen for remote procedure calls (RPCs) over TCP/IP from the USPAT control software for online processing and real-time visualization. It can also be used for offline processing and visualization.
Image processing algorithms are best implemented on specialized graphics hardware, such as the GPU, but in this study, we were able to achieve satisfactory performance with an optimized CPU implementation. The biggest performance gains came from substituting spatial domain algorithms with their frequency domain equivalents. Taking advantage of the Fast Fourier Transform, we can trivially improve the computational complexity of spatial filtering operations, which often have O(n2 ), time complexity, to O(n logn), which in practice is very close to linear time. Furthermore, for the filtering of raw RF data, we implemented fast discrete convolution with the Overlap-Add method18, which excels at finite impulse response (FIR) filtering.
Computation of the rHbT and %sO2
The computation of the functional parameters derived from the multispectral PAT data is implemented in the USPAT display software, and the functional parameters are automatically computed and visualized in real time. Briefly, we computed the oxy-hemoglobin and deoxy-hemoglobin (relative scale, rHbO and rHbR) concentration at every pixel by solving a non-negative linear least squares problem:
where g represents the measurements at four wavelengths, H represents the matrix of extinction coefficients of oxy-hemoglobin and deoxy-hemoglobin at each wavelength, and f represents the rHbO and rHbR. The rHbT is simply the sum of rHbO and rHbR, and the %sO2 can be computed from the ratio of rHbO:rHbT2. The computation of these parameters is implemented in the USPAT display software and is completely automated. This method with the system is validated through measuring the calibrated blood tube phantoms suspended in intralipid solution2.
USPAT control software
The USPAT control software automates the USPAT data acquisition process by communicating with the laser for wavelength tuning, the clinical US system for data acquisition, and the USPAT display software for data processing and visualization. After selecting the depth in the graphical user interface (GUI), the software sends a command to the US system (over TCP/IP through an ethernet cable) to load the correct sequence file. The Scan button begins the acquisition process of one set of coregistered multispectral PAT and US data. First, the control software sequentially tunes the laser wavelength (over USB) from the lowest to the highest, while the US system acquires the coregistered PAT and US frames. Finally, the control software triggers the USPAT display software (over TCP/IP) to compute the US and PAT B-mode images, reconstruct the functional maps, and display them in real time. At the same time, the laser is tuned back to the lowest wavelength.
Limitations
Currently, there are several limitations of the USPAT technique. First, photoacoustic imaging can reach only about 5 cm deep with commercial US transducers of 4-10 MHz bandwidth. Thus, for ovaries deeper than 5 cm, or when the target pathologic process is more than 5 cm from the vaginal fornix within a large adnexal mass, PAT is limited. Second, the limited field of view of the US transducer requires scanning a larger lesion at multiple angles to obtain an average that is more representative of the lesion's rHbT and %sO2 contrast. Third, the relative total hemoglobin concentration has been reported because the PAT measurements are the product of the local fluence distribution and optical absorption profile. It is challenging to estimate the optical absorption profile from in vivo measurements. Recently, neural network-based approaches have been explored for the reconstruction of the absolute total hemoglobin concentration19, but these approaches remain to be validated. Finally, the frame rate of multispectral photoacoustic imaging is limited by the speed at which the laser can tune its wavelength. The laser operates at 10 Hz and is mechanically tuned, and the data acquisition for four wavelengths takes about 15 s, so this is the bottleneck in improving the frame rate.
The authors have nothing to disclose.
This work was supported by the NCI (R01CA151570, R01CA237664). The authors thank the entire GYN oncology group led by Dr. Mathew Powell for helping with recruiting patients, radiologists Drs. Cary Siegel, William Middleton, and Malak Itnai for helping with the US studies, and the pathologist Dr. Ian Hagemann for helping with the pathology interpretation of the data. The authors gratefully acknowledge the efforts of Megan Luther and the GYN study coordinators in coordinating the study schedules, identifying patients for the study, and obtaining informed consent.
Clinical US imaging system | Alpinion Medical Systems | EC-12R | Fully programmable clinical US system |
Dielectric mirror | Thorlabs | BB1-E03 | Used to reflect light along the optical path |
Endocavity US transducer | Alpinion Medical Systems | EC3-10 | Transvaginal ultrasound probe |
Laser power meter | Coherent | LabMax TOP | Used to measure laser energy |
Multi-mode optical fiber | Thorlabs | FP1000ERT | Couple laser light to the endocavity ultrasound probe |
Non-polarizing beam splitter plate | Thorlabs | BSW11 | For splitting laser beam into sensors to measure energy |
Plano-concave lens | Thorlabs | LC1715 | For laser beam expansion |
Plano-convex lens | Thorlabs | LA1484-B | For laser beam collimation |
Plano-convex lens | Thorlabs | LA1433-B | Used to focus light into four optical fibers |
Polarizing beam splitter cube | Thorlabs | PBS252 | For splitting laser beam into four beams |
Protective probe shealth | Custom 3D printed | Hold and protect the four optical fibers at the tip of the ultrasound probe | |
Right angle prism mirror | Thorlabs | MRA25-E03 | Used to reflect light along the optical path |
Tunable laser system | Symphotic TII | LS-2145-LT50PC | Light source for multispectral PAT |
USPAT control software | Custom developed in C++ | Controls acquisition parameters of the ultrasound machine and the laser wavelength | |
USPAT image display software | Custom developed in C++ | Displays the US/PAT B-scans and sO2/rHbT maps in real time |