A review of co-registered transvaginal photoacoustic and ultrasound imaging for ovarian cancer diagnosis.

From Feb. 2017 to May 2018, 40 patients were enrolled in a pilot study at the Washington University School of Medicine and the Siteman Cancer Center [ 39 – 41 ]. These patients were clinically at risk for ovarian cancer or had an ovarian or pelvic lesion suggestive of malignancy. Prior to imaging with the PAI/US system, all patients were imaged with a commercial transvaginal US system (GE LOGIQ S8) by two radiologists who assigned a score (1–5) based on the standard of care imaging: 1 and 2, normal; 3, likely benign; 4, suspicious for malignancy; and 5, highly suspicious for malignancy. After the suspicious ovarian or pelvic lesion was examined, the commercial probe was withdrawn, and the customized PAI/US probe was inserted transvaginally to image the suspicious lesion. Because all the radiologists were familiar with the GE system, this two-step procedure facilitated a smooth transition from the GE system to the Alpinion system. The GE system also has better spatial resolution and depth than the Alpinion system in US mode. For each imaging location, three PAI and one US imaging frames were recorded at each wavelength. The total acquisition time for four wavelengths at each location was about 12–14 s. The overall time for several locations was about few minutes depending on number of ovaries/lesions. A first manuscript, presenting the initial results for 20 patients, was published in Radiology [ 39 ], and a second manuscript, with the results of all 40 patients was published in the Journal of Biophotonics [ 41 ]. An example of co-registered PAI and US imaging is shown in Figure 2 . The patient was a 50-year-old premenopausal woman with bilateral multicystic adnexal masses with septations and mural nodularity revealed by contrast-enhanced CT ( Figure 2(a) ). A 3.6 cm right adnexal mural nodule is shown as a solid mass, indicated by an arrow in Figure 2(a) . US Doppler indicated minimal blood flow in the solid area ( Figure 2(b) ). Note that color Doppler assessment of the lesion has been shown to be useful in the evaluation of malignancy, however, the reported accuracy ranges from 35% to 85% [ 42 ]. Figure 2(c) is an US image of the right adnexa, and (d) is a co-registered US and PAI rHbT map shown in color, with extensive diffused vascular distribution inside the ROI in the depth range of 1.5–4.5 cm, next to big cystic areas identified by US. Figure 2(e) is a CD31 stained histology image in the suture area, exhibiting dense microvessels; (f) is an %SO 2 map of the ROI marked by the white rectangular box in (d). Pathology showed well-differentiated FIGO stage I [ 43 ] endometrioid adenocarcinomas of both the right and left ovaries, which measured 8.3 cm and 20 cm, respectively. The mean rHbT measured in the ROI was 12.18 (a.u.), and the mean %SO 2 was 50.14%. Of the first 20 patients [ 39 ], 16 patients (mean age 51 years, range 34–68 years) with 26 ovaries were successfully imaged with the PAI/US system. Diagnoses ascertained by subsequent surgical pathology examination revealed high-grade serous carcinoma (n = 6 ovaries), endometrioid adenocarcinoma (n = 3), non-invasive serous borderline tumor (n = 2), Sertoli–Leydig cell tumor (a sex cord-stromal tumor; n = 1), normal ovaries (n = 5), and other causes of benign but enlarged ovaries (n = 9). The patients were grouped into invasive epithelial ovarian cancers (n = 9), other neoplasms (n = 3), and benign/normal ovaries (n = 14). The category of “other neoplasms” included the borderline tumors and Sertoli–Leydig cell tumor, with the rationale that both of these diagnoses have some potential for subsequent malignant behavior. Benign causes of enlargement included cystadenomas, endometriosis, a fibrothecoma, and other benign processes. As shown in Figure 3 , the average rHbT concentration computed from the four optical wavelengths was 1.9 times higher for invasive epithelial cancers than for the benign or normal ovaries (p = 0.01), with a difference of 5.4 (a.u.) (95% CI: 1.6 a.u., 9.2 a.u.%). Interestingly, the average rHbT of the two borderline serous tumors and one stromal tumor (the “other” group) was in the same range as that of the benign/normal ovaries, and it was not statistically different from the value for the benign/normal group (p = 0.56). However, the average %SO 2 of other types of tumors was in the same range as that of the invasive cancer group and was statistically different from the benign/normal group (p = 0.01). When the invasive epithelial cancers and other tumors were grouped together, the average %SO 2 of the combined group was 8.2% lower (95% CI: 3.2%, 13.2%) in comparison with the benign/normal group (p = 0.003). A t test (p = 0.192) showed that the CA125 value also did not differ between the invasive epithelial cancers, “other” tumor group, and the benign/normal group. However, because of skewed measurements, it was significant between the two groups with the Mann–Whitney test (p = 0.03). Note that the average rHbT and %SO 2 were computed for each patient, except for one patient who had one malignant and one normal ovary. In a second study of the data from all 40 patients, including the early 20 patients, we explored % SO 2 histogram features (mean, standard deviation, skewness, kurtosis, energy, and entropy) in addition to rHbT and %SO 2 in classification of ovarian cancer from benign lesions [ 40 , 41 ]. Among these features, three (mean, skewness, and energy) showed significant differences between the benign and malignant ovarian groups. Additionally, we also investigated the spectra of the PAI beam-formed data, which contain useful microstructural information of the imaged tissue [ 31 ]. To calculate the spectral features, the spectrum of each beamline in the ROI was calculated using the fast Fourier transformation (FFT). The average spectrum was then calculated by taking the average of the spectra of all the beamlines in the ROI. Then a line was fitted to this average spectrum. The slope of this line (SS), its intercept with 0.5 MHz frequency line (SI), and its value in the middle of the frequency range (MBF) provided useful information about microscale particles in the ROI. In Figure 4 , the rHbT and %SO 2 maps and PAI spectral features are compared for a malignant (a–d) and a benign (e–h) ovarian mass [ 41 ]. In these figures, the lesion regions are indicated by dashed rectangles in the overlaid rHbT (color-scale image) and background US (gray-scale) images. The malignant ovary has a stronger and more concentrated rHbT map than the benign mass. Histograms for each %SO 2 map are shown below in Figure 4 (c) and (g) . As can be seen, the mean %SO 2 is lower for the malignant group. Also, the histogram of the malignant ovary is skewed toward the lower values, while the benign ovary’s histogram is skewed towards the higher values. Finally, in Figure 4 (d) and (h) , the mean spectra of the PAI beamlines in the ROI, along with their fitted lines, are presented for a malignant and a benign ovary, respectively. The malignant ovaries show a smaller value of SS (more negative) and a larger value of SI (less negative). As demonstrated in the study by Amidi et al. [ 40 ], the lower SS in the malignant ovaries is related to the larger size of the absorbers in this type of masses, and the higher SI is associated with the larger size of the absorbers as well as a higher concentration of them in malignant ovaries. To use multiple parameters for ovarian cancer diagnosis, we developed support vector machine (SVM) classifiers to distinguish between the “malignant” and “benign/normal” groups. Data set from a total of 49 ovarian lesions of 40 patients was randomly divided into two groups. The first group included two-thirds of the data used for training, and the remaining one-third of the data was employed for testing the classifiers. To lower the chance of overfitting, this process was repeated 100 times. The SVM model with combined features (radiologist score, rHbT, %SO 2 mean, and SI 730) achieved a superior AUC value of 0.92 (95% CI: 0.89–0.95) on the testing data set. The sensitivity and specificity were 88% and 82%, respectively. The coregistered dual-modality photoacoustic and US technology meets the requirements of low-cost and non-invasiveness, and is easily incorporated into the standard of care, TUS. The initial results distinguishing between normal/benign ovarian tissue and ovarian cancer are encouraging. However, the photoacoustic imaging technology still faces challenges. First, human ovaries are complex organs that undergo cyclic changes in premenopausal women. There is variability within individual ovaries when compared to normal postmenopausal ovaries. To determine the variation of rHbT and %SO 2 , as well as spectral parameters, across the menstrual cycle, a group of premenopausal women could be monitored over a menstrual cycle at the follicular phase and the luteal phase. Second, certain benign ovarian lesions, such as endometriosis and inflammatory changes, can present similar rHbTand %SO 2 patterns to those of ovarian cancers. In the near future, we envision technology advancements and clinical validations. First we need to optimize the system and user interface and provide near real-time dual-mode US and PAI displays for radiologists to diagnose ovarian masses. Currently, we are using a clinical US system from Alpinion Medical Systems which is synchronized with the laser system for photoacoustic imaging. In principle, photoacoustic laser system can be integrated with any commercial US system by collaborating with US companies. The data acquisition of a complete data set of four optical wavelengths took about 12–14 s. This data acquisition time is limited by the mechanical tuning of the optical wavelength and can be improved by using acousto-optic tunable filters. The rHbT and %SO 2 maps can be computed immediately after data are transferred to a graphics processing unit (GPU) to facilitate near real-time diagnosis. Second, we need to validate the feature-based prediction models with a large patient cohort for more accurate ovarian cancer diagnosis. Currently, we are recruiting approximately 200 patients who are scheduled for surgery at the Washington University School of Medicine and Siteman Cancer Center. The second arm of the on-going clinical study explores the potential of early ovarian cancer detection and diagnosis by using the dual-mode photoacoustic and US technique to follow up a group of high risk young patients with genetic mutations. The successful completion of the project may optimize the clinical management of low suspicion benign ovarian tissue abnormalities by reducing surgery recommendations without compromising cancer detection, thereby lowering morbidity and health care cost. The longitudinal monitoring of a group of high-risk patients may lead to initiation of an effective screening study for early ovarian cancer detection and diagnosis, with consequently much improved overall survival of patients from this deadly disease. From the commercialization perspective, the first commercial dual-modality PAI/US system, developed by Seno Medical Instrument, has been approved by the FDA for breast cancer detection and diagnosis. We envision that our PAI/US system based on the same principle will follow the same path once the on-going clinical study demonstrates the efficacy of the dual-mode PAI/US technique in ovarian cancer diagnosis.

Essentially, any screening test should be cost-effective and easily incorporated into the standard of care in medical practice. An ideal test would be non-invasive and able to distinguish between normal/benign ovarian tissue and early stage ovarian cancer. Photoacoustic imaging (PAI) has advanced remarkably, and now provide spatial resolution and functional information at depths ranging from several millimeters up to several centimeters [ 21 – 25 ]. PAI is a hybrid imaging technology that uses nanosecond laser pulses to excite tissue. The laser induces photon absorption and thermoelastic expansion due to transient temperature rises that produce photoacoustic or ultrasound waves. Thus, PAI reveals the optical contrast of biological tissue with ultrasound resolution. Significantly, physiological parameters, such as relative total hemoglobin concentration (rHbT) and blood oxygen saturation (%SO 2 ), can be computed with PAI. rHbT is related to microvessel networks and thus to tumor angiogenesis, a key process for tumor growth and metastasis [ 26 , 27 ]. %SO 2 is an important indicator of tumor metabolism and therapeutic response [ 28 , 29 ]. The penetration depth of PAI is tunable with ultrasound frequency. In the diagnostic ultrasound frequency range of 3–10 MHz the penetration depth in tissue can reach 5 cm or more in the near-infrared spectrum. This penetration depth is adequate for transvaginal imaging of most ovarian lesions. Because the same US transducer can be used for both TUS and PAI detection, PAI is an ideal modality for enhancing TUS for ovarian cancer diagnosis. Our limited pilot data so far suggest that PAI parameters of rHbT and % SO 2 can improve TUS on identifying earlier stage ovarian cancer and accurate diagnosis of large solid benign masses. When rHbT and %SO 2 combined with TUS, a sensitivity of 88% and a specificity of 82% can be achieved. Further improvements are envisioned when more functional parameters derived from PAI and biomarkers are used together to develop diagnostic models.

Current markers and tools used for clinical diagnosis of ovarian cancer are serum cancer antigen 125 (CA 125), human epididymis protein 4 (HE4), and transvaginal ultrasonography (TUS). The routine use of CA 125 alone is not adequate for differential diagnosis, because other conditions, such as endometriosis, can also cause it to increase [ 9 , 10 ]. A number of large prospective studies reported that CA 125 and TUS are not sensitive and specific enough for early ovarian cancer diagnosis, detecting only 30–45% of ovarian cancer in the early stages [ 9 , 11 – 14 ]. The advanced imaging techniques of computed tomography (CT), magnetic resonance imaging (MRI), diffusion-weighted MRI (DWI-MR), and positron emission tomography ( 18 FDG-PET), are not recommended for detection of primary ovarian cancer [ 15 ]. Because most females with ovarian cancer experience non-specific symptoms, CT is often used to search for a cause of non-specific symptoms or to evaluate the abdomen after worrisome ultrasound findings. On CT imaging, ovarian cancer typically presents as thick-walled cysts with septations. In the pretreatment evaluation of ovarian cancer, CT has been considered the best available technique for determining the extent of disease (lymph nodes, cancer metastases) and assessing the likelihood of optimal surgical treatment [ 15 , 16 ]. The characteristics of ovarian cancer found by MRI imaging are partly similar to those found by CT: cystic lesions with septa and solid components. Additional features of malignancy revealed by MRI include septa thickness, nodularity, papillary projections, and necrosis [ 17 ]. Dynamic contrast-enhanced MRI and diffusion-weighted MRI can be useful second-line tools after ultrasonography, further differentiating between benign, malignant, and borderline lesions [ 16 ]. Currently, two-multicenter clinical trials are on-going to define the role of MRI in patients with advanced ovarian cancer [ 18 , 19 ]. Recently, guidelines on the role of 18 FDG PET/CT in diagnosis, staging, prognosis, therapy assessment, and restaging of ovarian cancer were published by the European Association of Nuclear Medicine and endorsed by the American College of Nuclear Medicine, the Society of Nuclear Medicine and Molecular Imaging, and the international Atomic Energy Agency [ 20 ]. 18 FDG PET/CT is recommended as the most useful modality for relapse detection and the second most useful for prognostic evaluation. There is less evidence of its value for therapy assessment, and there is very scarce or low quality evidence for its value in initial diagnosis and staging in patients presenting with a pelvic mass, and in treatment planning.

Ovarian cancer remains the deadliest of all the gynecological malignancies, with a 5-year overall survival rate between 35 and 49%; at the 10-year mark, the survival rate is approximately 24% [ 1 , 2 ]. When ovarian cancer is detected at an early, localized stage (stage I or II), surgery and chemotherapy can cure 70–90% of patients, but diagnosis at later stages (stage III or IV) yields a cure rate of 20% or less [ 3 ]. This poor prognosis is the result of a lack of early symptoms and an effective strategy for screening asymptomatic populations. A recent update from the U.S. Preventive Services Task Force concluded that major trials of promising ovarian cancer screening tools among healthy average-risk women have failed to show a benefit for screening [ 4 ]. Moreover, screening can pose considerable false positive risks, including major surgical complications in women found to have no cancer [ 4 ]. Women with a screening abnormality will generally undergo prophylactic bilateral salpingectomy or salpingo-oophorectomy, however, these procedures are not appropriate for normal risk women, who represent 75–85% of ovarian cancer cases [ 5 – 7 ]. Prophylactic oophorectomy results in morbidity and mortality as a consequence of premature menopause, including accelerated bone loss and cardiovascular death [ 8 ]. Clearly, early detection is critical, but the lack of effective screening tools causes only 20–25% of ovarian cancers to be diagnosed early. New diagnostic and screening methods are urgently needed to effectively evaluate the ovary.

The vaginal wall is typically less than 1 cm thick. In the anteverted uterus, the ovaries are positioned at the level of the cervix and slightly posterior, and are easily accessed with a TUS probe. Because of the reduced penetration depth, US attenuation is significantly reduced by a transvaginal approach, as opposed to an abdominal approach. For the same reason, transvaginal photoacoustic imaging is superior because the reduced tissue path minimizes both the light scattering from the laser beam to the imaged ovary and the photoacoustic wave attenuation from the imaged ovary to the US transducer. Our group has pioneered co-registered transvaginal photoacoustic/US imaging to evaluate ovarian lesions [ 30 – 41 ]. By simulating the light fluence and power output for different design parameters, we optimized the combined PAI/US probe for the highest light delivery output and best beam uniformity on the tissue surface [ 34 ]. The final probe was 3D printed, and the laser fluence profiles were experimentally measured through chicken breast tissue and intralipid solution at various imaging depths. A blood tube was successfully imaged below several centimeters of porcine vaginal tissue. This imaging depth was achieved with a laser fluence on the tissue surface of 20 mJ/cm 2 , which is below the maximum permissible exposure (MPE) in the near infrared wavelength range recommended by the American National Standards Institute (ANSI) [ 44 ]. Furthermore, the imaging capability was verified in ex vivo benign and malignant human ovarian lesions [ 30 , 31 , 34 , 35 ]. In addition to optimizing the probe design, we also designed a novel lens-array based illumination set-up for the compact co-registered PAI/US transvaginal probe [ 37 ]. The lens array consists of four cylindrical lenses that couple the laser beams into four 1-mm-core multi-mode optical fibers, with an optical coupling efficiency of ~70% ( Figure 1 ). The illumination fibers are then affixed to a PAI/US probe. A clinical US system (Alpinion Medical Systems, Republic of Korea) with a transvaginal probe (EC-12R) was used for our clinical study [ 39 – 41 ]. The system consists of 1) a fully programmable clinical US system, 2) a customized optical fiber-based light delivery system coupled with the EC-12R transvaginal US probe, and 3) a Nd:YAG laser pumping a pulsed, tunable (690–900 nm) Ti-sapphire laser. A time-division multiplexing approach was used during co-registered mode, wherein each PAI frame was synchronized with one laser pulse, and three consecutive PAI frames were acquired for averaging. Each US frame was recorded at each of four optical wavelengths (730 nm, 780 nm, 800 nm, and 830 nm) during in vivo imaging. These wavelengths were selected based on the absorption properties of oxygenated and deoxygenated hemoglobin, which are the main chromophores related to tumor angiogenesis and oxygen microenvironment. The control system was customized in Python, and the laser wavelength synchronization with the PAI/US system was done in C++. The standard delay-and-sum beamforming algorithm was used for US imaging. For PAI, the delay-and-sum was used to form an image of each wavelength and images from four optical wavelengths were weighted by absorption extinction coefficients to compute relative oxygenated and deoxygenated hemoglobin concentrations (oxyHb, deoxyHb). The relative total hemoglobin image (rHbT) is the summation of the oxyHb and deoxyHb and the blood oxygen saturation map (%SO 2 ) is the ratio of oxyHb over rHbT [ 39 ]. The transvaginal PAI/US probe consisted of a 128-element array transducer with a 6 MHz central frequency and 80% bandwidth, surrounded by four 1-mm core diameter multimode fibers for light delivery. A custom 3D-printed sheath enclosed both the fibers and the transducer for patient studies.