Characterization of adnexal lesions using photoacoustic imaging to improve sonographic O-RADS risk assessment.

The study protocol was approved by the IRB of Washington University School of Medicine and was HIPAA compliant and was registered with ClinicalTrials.gov (Identifier: NCT04178018 ). This study enrolled patients referred to an academic gynecologic oncology practice that had planned oophorectomy for an ovarian/adnexal lesion(s) suggestive of malignancy. Patients were identified by the GYN physicians and consented by the study coordinator prior to the date of the scheduled surgical procedure. A total of 73 patients consented and participated in this study from August 2020 to June 2022. Out of these patients, five were not included in the data analysis: one patient had no ovarian/adnexal lesions visualized by clinical US imaging, one patient did not undergo surgery, one patient had a deep lesion greater than 5 cm away from the transducer (non-reliable PAI imaging), and two patients’ lesions were not adequately scanned due to technical problems ( Figure 1 ). Thus, the total number of patients was 68 with an average age of 52 years ranging from 21 to 79 years, and the total number of imaged and evaluated ovarian/adnexal lesions was 93. In O-RADS assessment, the 93 ovarian/adnexal lesions were categorized into 7 histologic groups of 1) malignant ovarian/adnexal lesions; 2) malignant fallopian tubes only; 3) benign solid lesions (fibrothecoma, fibroma, leiomyoma and benign Brenner tumor); 4) endometriosis, adhesions, abscess; 5) benign cystic lesions; 6) mature teratoma; and 7) normal ovaries ( Table 2 and 3 ). Notably, 12 patients each had bilateral lesions of the same categories, and 13 patients had bilateral lesions of different categories. For the 13 normal ovaries, 5 out of 13 were surgically removed with surgical pathology results. The remaining eight normal ovaries were left in place at the time of laparoscopy to preserve reproductive function in these young patients. Group demographics and detailed information on the included lesions are presented in Table 1 . In the statistical analysis and logistic regression models, only one lesion was selected per patient to ensure data independence. If O-RADS scores were different for patients with bilateral lesions, the lesion with a higher O-RADS category was selected for analysis. If the O-RADS categories were the same for patients with bilateral lesions, the larger lesion was selected for analysis. CT exams of the abdomen and pelvis were available for some patients before oophorectomy and were used to assist the study radiologists in assessing the location of any lesions relative to the vaginal apex. For each patient, the study radiologist assisted by a sonographer scanned the suspicious ovarian/ adnexal lesion(s) using a clinical US system (LOGIQ S8, General Electric, Boston, MA, USA) and a transvaginal transducer (iC5–9-D). After the suspicious lesions were examined, the clinical probe was withdrawn and the customized PAI/US transvaginal probe was inserted to image the suspicious lesions. The US O-RADS assessment was based on clinical GE Logiq S8 system which has better spatial resolution and depth than the customized PAI/US system in US mode. Our customized PAI/US system and probe consisted of a fully programmable US system (EC-12R, Alpinion Medical Systems, Republic of Korea), a customized optical fiber-based light delivery system coupled with a transvaginal US probe (6 MHz, 80% bandwidth), and a Nd:YAG laser pumping a pulsed, tunable (690–890 nm) Ti-sapphire laser at 10 Hz (Symphotics, Camarillo, California). The laser energy density at the probe-tissue surface was maintained below the maximum permissible exposure (MPE) 19 . A time-division multiplexing approach was used during the co-registered mode, wherein each US B-scan was recorded followed by a PAI B-scan synchronized with and averaged from five laser pulses. Four optical wavelengths (750 nm, 780 nm, 800 nm, and 830 nm) were used for imaging. Total hemoglobin concentration (relative scale, rHbT) and blood oxygen saturation %sO2 were calculated from 4 wavelength data using extinction coefficients reported in the literature ( Appendix S1 ). The details of the custom PAI/US system and the study protocol can be found in references 7, 14 and 20. A region of interest (ROI) for each lesion was identified by the study radiologist based on the co-registered gray-scale US image obtained from the customized PAI/US system. In general, the entire suspected area was selected using a rectangular window as the ROI for PAI analysis. For complex cystic lesions, the descriptors in O-RADS 1 – 2 , such as unilocular or multilocular cysts with an irregular inner wall and/or irregular septation, or solid/solid appearing components, or papillary projections, were considered as risk factors for malignancy. Thus, the rectangular ROIs for large complex cystic lesions were selected from these suspicious areas identified by the study radiologist. A mean value inside the full width at half maximum (FWHM) of the rHbT map inside the ROI was computed to represent the lesion hemoglobin contrast and a mean %sO2 value of voxels with %sO2 greater than zero inside the ROI was computed to represent %sO2. For each lesion, 4 to 5 data sets were acquired and an average of mean rHbT and mean %sO2 were used to quantify each lesion. The rHbT FWHM depends on rHbT distribution and independent of the ROI size and has been found to be a reliable parameter for quantifying the rHbT for each lesion 7 , 14 . Because some ovarian/adnexal lesions were heterogeneous, our procedure required surgeons to tag the lesion with a suture at a site proximate to the vaginal apex, providing a fiducial marker to correlate the PAI/US imaged area with histologic examination. After the completion of imaging using the clinical US system, the study radiologist graded the lesions/ovaries using O-RADS 1 – 2 . A second study radiologist independently graded the lesions/ovaries at a later time based on saved gray scale and color Doppler images and cine clips acquired from the clinical US system. Four study radiologists with 33 (WM), 29(CS), 5 (MI), and 2 (MH) years of experience participated in the imaging and imaging assessment based on the O-RADS guideline. Each patient’s lesions/ovaries were assessed by one senior ( > 25 years experience) and one junior (≤5 years experience) study radiologist, and all scores were reported in Tables 2 and 3 . To evaluate reader agreement based on real-time US images (reader #1) and recorded US images (reader #2) obtained from the clinical US system, Fleiss’s Kappa coefficient (κ) was calculated for O-RADS and color Doppler scores of all lesions using the R package “irr” (Indianapolis, Indiana, USA). Additionally, the agreement of readers based on experience (>25 years) and (≤5 years) was assessed using Fleiss’s Kappa coefficient. The O-RADS US defines six categories for risk classification. These include O-RADS 0, an incomplete evaluation; O-RADS 1, physiologic findings (normal ovary); O-RADS 2, almost certainly benign (<1% risk of malignancy); O-RADS 3, lesions with low risk of malignancy (1% to 10%); O-RADS 4, lesions with intermediate risk of malignancy (10% to 50%); and O-RADS 5, lesions with a high risk of malignancy (>50%). Color Doppler US was evaluated using color score (CS) of 1–4 as defined per O-RADS with CS 1, no flow; 2 minimal flow, 3 moderate flow, and 4 very strong flow. For this study, normal ovaries were also graded with O-RADS and CSs for consistency. Note that the region of interest (ROI) used to assess the CS was based on the clinical US system and the standard clinical protocol. Because Doppler US images the blood flow in larger vessels, while PAI maps lesion vasculature and %sO2 distributions, we did not attempt to select exactly the same ROI for comparing color Doppler US and PAI. Additionally, because the clinical US system and the customized PAI/US system have different resolutions and imaging depths, selecting exactly the same ROI for each lesion is not feasible or needed for this study. Statistical analyses and logistic regression models for risk assessment were based on the most suspicious lesion per patient in 68 patients. One-way ANOVA test was used for comparison of cancer groups with different benign groups followed by a two-tailed post-hoc Dunnett test of significance 21 . Minitab 19 software (Minitab, State College, PA, USA) was used for statistical calculations. A corrected P value, Pc, less than 0.05 was considered statistically significant. A random forest model was employed to rank the importance of each predictor or feature (O-RADS, CS, maximal dimension based on US measurements, rHbT and %sO2), which was calculated by averaging the Gini impurity at each node of the forest where that predictor was used. Scikit-Learn, a module in Python 3.7, was used to calculate the feature’s importance 22 . Generalized logistic regression ( Appendix S2 ) was used to relate diagnostic outcomes to individual predictor variables for ROC analysis. Matlab R2020a (Natick, Massachusetts, USA) was used for logistic regression analysis. The ROC analysis and 95% confidence interval (CI) of each area under ROC (AUC) were computed in R using the “pROC” package.

In O-RADS assessment, the 93 ovarian/adnexal lesions were categorized in 7 groups of 1) malignant ovarian lesions (n=19); 2) malignant fallopian tube lesions (n=2); 3) benign solid lesions (fibrothecoma, fibroma, leiomyoma and benign Brenner tumor) (n=12); 4) endometriosis, adhesions, abscess (n=14); 5) benign cystic lesions (n=23); 6) mature teratoma (n=10); and 7) normal ovaries (n=13) ( Table 2 and 3 ). In statistical and regression model analysis, one lesion per patient was selected. There were a total of 14 patients with malignant ovarian lesions and 2 patients with malignant fallopian tube lesions, and 52 patients had benign lesions consisting of: 1) 10 patients with benign solid lesions (fibrothecoma, fibroma, leiomyoma, and benign tumor); 2) 8 patients with endometriosis, adhesions, abscess, 3) 23 patients with benign cystic lesions (serous or mucinous cystadenoma), 4) 9 patients with mature teratoma, and 5) 2 patients with normal ovaries. The distribution of O-RADS reads is shown in Table 2 . The reader agreement based on real-time US and recorded US images for O-RADS was good (κ = 0.69, p<0.001) and for CS was moderate (κ = 0.51, p<0.001). The agreement between junior and senior reader groups was good (κ = 0.70, p<0.001) 23 . The radiologists were sensitive in detecting malignant and borderline tumors as 100% (38/38) of these tumors were read as O-RADS 4 and 5. However, the 3 reads of O-RADS of two fallopian tube lesions with high-grade carcinoma ranged from O-RADS 1 (benign) to 4 (intermediate, 10–50% risk of malignancy). One reader (second read) could not identify the fallopian tube lesion with confidence from recorded US images. Additionally, a range of 7.1–58.3% of benign lesions was categorized as O-RADS 4 or 5. Specifically, 58.3% (14/24) of solid fibrothecoma/fibroma/leiomyomas/benign tumors, 7.1% (2/28) of solid endometriosis/adhesions/abscess, 37.0% (17/46) of cystadenoma, 10% (2/20) of mature teratoma, and 7.7% (2/26) of normal ovaries were read as O-RADS 4 or 5 (suspicious for malignancy). Furthermore, an additional 28.6% (8/28) of benign endometriosis/adhesions/abscess lesions, and 37.0% (17/46) of benign cystadenomas were read as O-RADS 3. The Doppler CS is considered a useful descriptor to characterize the blood flow of the lesions. As shown in Table 3 , 55.3% (21/38) of malignant lesions, and 25.0% (6/24) of solid fibrothecoma/fibroma/leiomyoma/benign tumors were read as CS of 3 or 4. Only 8.7% (4/46) of benign cystadenoma were read as CS of 3, and none had CS of 4. There is considerable overlap in CS between malignant and benign lesions with CSs 1 and 2. The rHbT of the malignant lesion group was significantly higher than those of all benign lesion groups with corrected Pc values given in Figure 2A . Statistical significance tests were not performed between the malignant groups and normal ovaries due to the small number in the normal group. The rHbTs of two malignant fallopian tube lesions were in the same category as the malignant ovarian lesion group after Dunnett pairwise comparison; however, there was no statistically significant difference between the malignant fallopian tube group and benign groups, except the mature teratoma group (Pc=0.049), due to the small number of malignant tubes. The mean values of rHbT of the malignant ovarian lesions, malignant fallopian tubes, solid benign solid lesions, endometriosis/adhesions/abscesses, benign cystic lesions, and mature teratomas were 9.9 (a.u), 8.7 (a.u), 4.6 (a.u), 5.7 (a.u), 5.9 (a.u) and 4.5 (a.u), respectively. The average of rHbT of the malignant lesion and fallopian tube groups was 1.8 times higher than that of benign groups. The %sO2 of the malignant lesion group was 5% lower than those of the benign lesion groups; however, statistical significance between benign and malignant lesion groups was not reached ( Figure 2B ) due to a large standard deviation in the malignant group. CA125 of patients with malignant lesions was significantly different than that of benign solid lesions, benign cystic lesions, and mature teratoma groups, but not significantly different than the endometriosis/adhesions/abscess group ( Figure 2C ). We considered whether various features could be combined to predict the risk of malignancy. Six features were entered into a logistic regression: serum CA125, maximal lesion dimension measured by US, O-RADS score, rHbT, %sO2, and CS. The first and second O-RADS and CS reads were averaged to obtain average O-RADS and CS for logistic regression models. The average O-RADS yielded an AUC of 0.86 as compared with first O-RADS reads of 0.84 and second reads of 0.82. Because each lesion was read by one senior and one junior radiologist, an average performed better by taking radiologists’ experience into account. The rank of importance of each predictor is shown in Figure 3A . rHbT was the most important predictor, followed by CA125, O-RADS, %sO2, CS, and lesion diameter. As shown in Figure 3B ROCs, the AUC of CA125 alone was 0.79 (95% CI: 0.626–0.953), the AUC of O-RADS was 0.86 (95% CI: 0.733–0.981), the addition of CA125 to O-RADS slightly improved the AUC from 0.86 to 0.88 (95% CI: 0.760–1), and the AUC of combined rHbT and %sO2 was 0.90 (95% CI: 0.815–0.985). Since the study radiologists already incorporated the lesion size and CS into the overall O-RADS score for each lesion, adding two additional imaging predictors of size and CS to O-RADS gave no improvement, with an AUC of 0.86 (95% CI: 0.737–0.990). The AUC of a model with rHbT alone was 0.87 (95% CI: 0.758–0.984) and the addition of %sO2 improved the AUC to 0.90. The addition of O-RADS to rHbT and %sO2 improved AUC from 0.90 to 0.93 (95% CI: 0.863–0.997), and the addition of CA125 to rHbT and %sO2 improved AUC from 0.90 to 0.94 ( 95% CI: 0.858–1). As shown in Figure 3B , the best AUC of 0.97 (95% CI: 0.932–1) was achieved when four predictors (CA125, O-RADS, rHbT, and %sO2) were used. Also shown in Figure 3B , O-RADS, CA125 and O-RADS, and rHbT and %sO2 achieved a sensitivity of 93.3%, however, the specificity of rHbT and %sO2 was 78.6% which was 14.3% higher than that of O-RADS only at 62.5%, and 3% higher than that of CA125 and O-RADS. The best sensitivity of 100% and specificity of 82% was achieved with four predictors of CA125, O-RADS, rHbT, and %sO2. The logistic regression models are given in Appendix S2 . Examples of malignant ovarian lesions ( Figures 4 – 5 ), a malignant fallopian tube ( Figure 6 ), and benign ovarian lesions of various pathology ( Figures 7 –10) are summarized here. As shown in Figure 4 , the rHbT level (red color in D) superimposed on the co-registered US was high with a mean value of 8.5 (a.u) and the spatial and depth distribution was intense and diffuse (high rHbT level distributed over a large spatial and depth extent) in the malignant mesothelioma. The %sO2 map (color in E) showed a low level of mean %sO2 of 56.4%. The O-RADS scores from the two study radiologists were 5 and 4. Figure 5 shows an example of stage I high-grade serous carcinoma of the right ovarian lesion. The rHbT map (C) superimposed on the co-registered US showed a high mean signal level of 8.9 (a.u) with a diffuse distribution around the lesion. A %sO2 map (D) showed a high mean level of %sO2 of 72.3%. The O-RADS scores from the two study radiologists were 4. An example of a high-grade serous carcinoma of the right fallopian tube and a normal right ovary is shown in Figure 6 . The relative total hemoglobin map showed a high hemoglobin content of 7.2 (a.u) in the fallopian tube area (D) and the %sO2 showed a moderate level of 65.2% (E). The two study radiologists graded the lesion as O-RADS 3 and 4. Among benign ovarian lesions of various histologies, a large fibroma ( Figure 7 ) was graded as O-RADS 4 and 5 due to its solid appearance and suspicious lobulated posterior margin; however, it showed scattered and extremely low hemoglobin contrast of 2.8 (a.u) (D). %sO2 map was not computed due to low hemoglobin content. A large unilocular cystic lesion with at least 4 papillary projections ( Figure 8 ) was graded as ORADS 4 and 5 but showed scattered and low hemoglobin contrast of 4.1 (a.u) (C), and a moderate %sO2 level at 62.6% (D). A large intraovarian multilocular cyst with a solid component or a papillary projection was graded as ORADS 3 and 4 ( Figure 9 ). A PAI rHbT map superimposed on co-registered US showed diffuse signals in the solid area at a low level of 5.6 (a.u) (C), and %sO2 map showed moderate level of 65.4% (D). The last example is a 21-year-old woman with a right adnexal lesion of right ovary with an anechoic cystic component inside and an additional solid appearing lesion with hyperechoic component and posterior acoustic shadowing ( Figure 10 ). The lesion was graded as O-RADS 2 by both radiologists. A co-registered US focused on the right ovary is shown in B. An rHbT map (C) superimposed on co-registered US showed low level of 3.9 (a.u), and a %sO2 map (D) showed moderately higher level at 66.5%. Pathologic examination showed a mature teratoma.

Recent guidelines based on MRI and US O-RADS have recommended that radiologists standardize the imaging assessment of adnexal/ovarian masses for clinical risk management and surgical triage 1 – 2 , 24 – 25 . Since most adnexal lesions are first detected with US, US O-RADS is considered the primary assessment tool and is accurate in identifying simple cysts and classic benign lesions such as hemorrhagic cysts, endometriomas, and dermoid cysts 24 . However, US is often limited for definitive characterization and risk stratification of other types of ovarian lesions, including solid appearing lesions or components of cysts with or without blood flow. As a result, most of these lesions are surgically resected and ultimately found to be benign, and patients may be exposed to personal and economic costs related to unnecessary oophorectomy. Recently, MRI has been recommended to further characterize sonographically indeterminate adnexal lesions, but it is relatively expensive 26 – 27 . Thus, additional accurate non-invasive methods of risk stratification of adnexal lesions are needed. We have demonstrated in this study, the rHbT level was high and it distribution was more intense and diffuse in large malignant lesions. The greatest value of PAI is to improve the diagnostic accuracy for large, solid, benign lesions and large cystadenomas with solid components, which typically showed low hemoglobin contrast as compared with malignant lesions. The most dramatic potential application of PAI is in identification of vascular contrast of smaller cancers, such as malignant fallopian tube and stage I cancers, thus allowing earlier detection. Given the low incidence of malignant fallopian tube and earlier-stage cancers in patients referred for oophorectomy, the potential role of PAI needs to be validated with a large screening patient cohort. Note that this study was completed before the publication of ACR O-RADS 2022 US v2022 28 . The updated O-RADS included terms of “shadowing of smooth solid lesions” signifying fibromatous lesions and “bilocular smooth cyst without a solid component” suggesting a downgrade in O-RADS. These terms add additional specificity to O-RADS to these categories. However, PAI can further downgrade O-RADS of these lesions and has a large role in the assessment of multilocular cysts. The literature data regarding the angiogenesis and %sO2 of human fallopian tube adenocarcinoma is scant. In addition, molecular differences that may exist between tubal and ovarian adenocarcinomas of similar histologic type have not been extensively characterized. In a related study, Kalir et al. compared glucose transporter GLUT1 in human malignant fallopian tube epithelia (all serous, n=29), and ovarian serous carcinomas (n=37) and found GLUT1 staining in primary fallopian tube cancers was less extensive than in primary ovarian adenocarcinomas 29 . Because the frequent localization of GLUT1 positivity to regions most distal from stroma/stromal capillaries is known to cause activation of GLUT1 expression by hypoxia-sensing cellular pathways 29 – 30 , this result may suggest better blood perfusion and oxygen level in fallopian tube cancers than in ovarian adenocarcinomas. In our pilot study 7 , 14 , there were a total of 7 epithelial ovarian cancers (EOC) and 1 sex cord-stromal tumor. Based on analysis of one lesion per patient, the %sO2 was 8.2% lower than that of the benign and normal group and this difference was statistically significant. Compared with these earlier data, the patient cohort in the current study had a wider spectrum of cancers. The tumor angiogenesis and hypoxia of EOC, the most lethal gynecological cancer, have been reported in the literature 31 . The process of angiogenesis is crucial in the development and progression of EOC. As tumors grow and become less oxygenated and more hypoxic, cancer cells develop mechanisms to survive under lower oxygen tension 32 . This development may explain the higher %sO2 of the only stage I high-grade serous carcinoma in this study ( Figure 5 ) as well as that/those in the prior study. In EOC, women with advanced disease (stage III or IV) have a specific hypoxia inducible factor (HIF) profile composed of elevated nuclear expression of HIF1α and elevated cytoplasmic HIF-2α expression 33 . This specific profile has been shown to be associated with a poor prognostic outcome, and PAI %sO2 may be significant in predicting the effect of ovarian cancer therapies 34 . The two metastatic cases showed a higher %sO2 within the malignant group. We do not have an explanation, particularly given the limited number of cases. One large mucinous adenocarcinoma (14 cm) did not show any hemoglobin contrast. A review of H&E of the surgical specimen showed that the lesion had 15% invasive mucinous adenocarcinoma arising in a background of 85% borderline mucinous tumor and it is not certain that the malignant component underwent PAI. Due to its solid appearance and large size, the lesion was scored O-RADS 4 and 5. The ovarian cystic lesions referred for oophorectomy were complex. In general, cystic lesions do not generate much photoacoustic signal inside the liquid area, however, scattered signals of low hemoglobin contrast can be present in the boundary area with mixed solid and cystic components, especially from cyst fluid of serous cystadenomas. The cyst fluid from serous cystadenomas is non-viscous and may contain blood because of the nature of the lesions 35 . Two outliers with high rHbT in the benign cystic lesion group were serous cystadenomas. The PAI signals from these two cystic lesions filled the entire cyst as delineated by the co-registered US. This distribution is quite different from the high hemoglobin contrast distributed at the lesion boundaries and inside in malignant lesions. Future work includes using pattern recognition via neural networks to identify features for improving the diagnosis of cystadenomas 36 – 38 . PAI/US technology has limitations. First, lesions deeper than 5 cm cannot be imaged with a good signal-to-noise ratio with our customized PAI/US system. 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 hemoglobin and %sO2 contrast. Third, the relative total hemoglobin concentration has been reported due to challenges to separate the optical absorption from the in vivo measurements ( Appendix S1 ). Recently, neural network-based approaches have been explored for the reconstruction of absolute total hemoglobin concentration 38 which remain to be validated from patient data. In summary, data from 68 patients with adnexal/ovarian lesions revealed that rHbT, O-RADS, CA125, and %sO2 are predictors of malignancy. A model utilizing all four features achieved superior performance with AUC=0.97 (95% CI: 0.932–1), and accordingly the best sensitivity of 100% and specificity of 82%. The additional PAI physiological parameters of rHbT and %sO2 significantly improve US O-RADS in the accurate diagnosis of ovarian/adnexal lesions.

Malignant and benign ovarian/adnexal lesions represent a heterogeneous spectrum of disease with overlapping ultrasound (US) imaging characteristics. The Ovarian-Adnexal Reporting and Data System (O-RADS) US risk stratification and management system provides guidelines for US reporting, allows assignment of risk to ovarian and other adnexal masses, and has associated management recommendations for each risk category 1 . The goals of O-RADS are to optimize ovarian cancer outcomes, and to minimize unnecessary surgical procedures for patients with low malignancy risk 2 . Thus, O-RADS serves to improve preoperative triage via accurate risk assessment, which, in turn, reduces health care costs and surgical complications. Gray scale and color Doppler ultrasound are central to the O-RADS paradigm, but photoacoustic imaging (PAI), which measures functional parameters in real time, has great potential to be included in this model. PAI illuminates with near-infrared light at specific wavelengths selectively absorbed by oxygenated and deoxygenated hemoglobin; the photoacoustic wave generated from the light absorption can “map out” hemoglobin contrast 3 – 4 , and this can be registered to anatomical structures. Blood oxygen saturation, %sO2, can be computed as a ratio of oxygenated hemoglobin to total hemoglobin. Both total hemoglobin concentration and %sO2 provide functional information on tissue vascularity and oxygen consumption 5 – 6 . Since malignant lesions generally have higher rHbT and lower %sO2 7 , these two independent variables may help differentiate malignant from benign ovarian/adnexal lesions and improve surgical management recommendations based on O-RADS assessments. Recent PAI oncology applications include breast cancer 8 – 10 , skin cancer 11 , thyroid cancer 12 – 13 , ovarian cancer 7 , 14 , cervical cancer 15 , prostate cancers 16 – 17 , and colorectal cancer 18 . PAI in the setting of adnexal malignancy risk stratification, to our knowledge, is largely unexplored. Since the US transducer is used for receiving photoacoustic waves, PAI co-registered with US provides complementary diagnostic data involving structure and function. The co-registered US localizes the lesion, and the PAI images inform tumor hemoglobin content and %sO2 distribution. Co-registered PAI/US has shown promise in a pilot study for differentiating malignant from benign lesions 7 , 14 . It is possible that information obtained from PAI/US could be used to refine O-RADS assessments to improve patient risk management recommendations. This study aimed to characterize the rHbT and %sO2 distributions of lesions in different risk categories and assess the diagnostic impact of PAI when combined with US O-RADS assessment in women with ovarian/adnexal lesions.

Appendix S1: Computation of relative total hemoglobin and %sO2 calculations Appendix S2: Logistic regression model