Functional photoacoustic/ultrasound imaging for the assessment of breast intraductal lesions: preliminary clinical findings

: This study aimed to identify features of breast intraductal lesions in photoacous-tic/ultrasound (PA/US) imaging and compare PA/US with color Doppler flow/ultrasound (CDFI/US) in the evaluation of breast intraductal lesions. In the nine patients with 10 breast intraductal lesions and 8 patients with 8 benign lesions, total vessel scores evaluated from PA/US are significantly greater than those from CDFI/US (p = 0.005). PA internal vessel scores and oxygen saturation (SO 2 ) score are significantly increased in breast intraductal lesions than in benign lesions (p = 0.016, p = 0.006). With a cutoff PA score (sum of PA internal vessel score and SO 2 score) of 2.5, we obtained a sensitivity of 90% and a specificity of 87.5% in differentiation of two groups. PA/US upgraded 40% of breast intraductal lesions, and downgraded 50% of benign lesions from the Breast Imaging Reporting and Data System grading results based on CDFI/US. PA/US functional imaging has the potential in differentiating breast intraductal lesions.


Introduction
Breast intraductal lesions are a heterogenous group of diseases, including intraductal papilloma (IDP), atypical papilloma, papilloma with ductal carcinoma in situ (DCIS), and intraductal papillary carcinoma [1,2]. Although most papillary lesions are benign, women with papilloma/papillomatosis have 1.5 to 2 times greater chance of getting invasive breast carcinoma (IBC) than the general female population [3]. It was reported that 16.4% of papilloma diagnosed by core needle biopsy may coexist with DCIS and IBC after surgery [4]. As a consequence, rather than routine follow-up which is commonly suggested for other benign breast lesions, surgical excision is warranted in virtually all papilloma cases against the subsequent carcinoma [4][5][6][7].
Up to date, it is still challenging for radiologists to diagnose breast intraductal lesions by clinical imaging. In general, breast ultrasound (US) has a better sensitivity than X-ray mammography due to its ability of revealing branching ducts under the nipple, a typical feature of breast intraductal neoplasm [8]. Among all the intraductal lesions, solitary papilloma is associated with a higher risk of breast cancer [9] and may present similar with other benign lesions in US as a small, solid, hypoechoic, and heterogeneous mass [8,9]. Similarly, DCIS, another type of breast intraductal lesion of which 19% to 50% can develop into invasive breast cancer [10,11], also may not show typical US imaging features such as irregular shape, non-parallel orientation with respect to tissue layers as late stage breast cancers and thus being difficult to diagnose. These clinical challenging can lead to unnecessary biopsies and therefore, new imaging modality for better diagnosis of breast intraductal neoplasm especially for solitary papilloma and DCIS is highly desired.
As an emerging imaging modality, photoacoustic (PA) imaging (PAI) is based on the photoacoustic effect [12,13]. When pulsed light is absorbed by optical absorbers inside tissue and instantly converted into heat, a local pressure will rise and generate subsequent ultrasonic waves (PA signal). Previous clinical studies have validated the potential of PAI to differentiate benign breast masses from malignant ones due to its capability of revealing functional information such as oxygenation saturation (SO 2 ) [14][15][16][17][18]. PAI combined with traditional handheld ultrasound imaging has gained lots of interest recently for its operational friendly in clinical environment [15,18]. With real-time PA/US dual mode imaging, the operator can first locate and get morphological information of lesions with US imaging and then obtain complementary functional information provided by PAI in the same region. In addition, PAI based on endogenous photo absorbers such as hemoglobin is non-invasive. Based on the phenomenon that malignant tumors can trigger angiogenesis and outgrow its blood supply, which leaves a lower oxygenation saturation than normal tissues [19], PA/US functional imaging may have the potential to enhance US specificity in differentiating malignant tumors based on the abnormity of oxygenation saturation [20][21][22]. Nevertheless, very few studies have reported the study of functional PA/US in differentiating breast intraductal tumor from typical benign breast lesions such as fibroma or breast adenoma. In this work, we aim to identify the morphological and functional features of breast intraductal lesions with PA/US imaging and compare it with CDFI/US imaging.

Method
Our study is prospective. The study protocol (No. JS-1424) was approved by the institutional ethics committee of Peking Union Medical College Hospital. Written informed consent was obtained from all participants.

PA/US dual mode functional imaging system
The imaging system in this study is based on modification of a high-end clinical US machine (Resona 7, Mindray Bio-Medical Electronics Co., Ltd.). US, CDFI imaging could be displayed along with PA imaging, which is reconstructed and displayed online.
The handheld probe used in this study has 192 elements and a central frequency of 5.8 MHz (L9-3U, Mindray Bio-Medical Electronics Co., Ltd.). The -20 dB bandwidth of the transducer is 8 MHz, ranging from 1.8 to 9.8 MHz. The laser source for PAI is an OPO tunable laser (Spitlight 600-OPO, Innolas laser GmbH), which generates 700-850 nm laser pulses at 10 Hz. In our study, we used 750 nm and 830 nm for PA imaging. We chose these two wavelengths for two reasons. First, deoxygenated hemoglobin has stronger absorption at 750 nm, and oxygenated hemoglobin has stronger absorption at 830 nm. Second, the pulse energy of the OPO laser maintains high at these two wavelengths, which could benefit the overall signal to noise ratio. Laser was delivered through a one-two bifurcate optical fiber bundle mounted along two lateral sides of the probe ( Fig. 1). At each wavelength, one frame was generated per pulse and reconstructed with the delay and sum algorithm for real time PAI. The overall combined wavelength switching, data acquisition, and PAI reconstruction time was around 10 ms. Therefore, given the PRF (10 Hz) of the laser system, PA/US real time imaging has a frame rate of 10 Hz for single wavelength and 5 Hz for two wavelengths when acquiring SO 2 maps. Before calculating SO 2 , PA signals at different wavelengths were normalized by the pulse energy recorded in real time. To exclude noise signals from PA images and SO 2 maps, for all PA images, we applied a threshold of 35% of the maximum pixel value. PA signals below the threshold were not shown in PA images and were not used for calculation of SO 2 maps.

SO 2 analysis
In the wavelength range from 700 nm to 900 nm, the dominant endogenous photo-absorber is hemoglobin, which can be either oxygenated hemoglobin (Hb) or deoxygenated hemoglobin (deHb) [23]. PA signal with optical wavelength λ i at the spatial position r can then be described as Eq. (1) [24]. Γ is the Grüneisen parameter of the tissue. F(λ i , r) is the optical fluence with optical wavelength λ i at the spatial position r. C(Hb, r) and C(deHb, r) are the concentrations of Hb and deHb, respectively. ε Hb (λ i ), ε deHb (λ i ) are the known molar extinction coefficients of Hb and deHb at wavelength λ i . The parameter k is a proportional factor determined by the characteristic of the imaging system.
Because both the absorption coefficient µ a (λ) and the reduced scattering coefficient µ ′ s of the background breast tissue at 750 nm and 830 nm are similar [25,26], the optical fluence F(λ 750nm , r) and F(λ 830nm , r) were assumed to be approximately same after normalization with laser illumination power at each wavelength. The definition of oxygenation saturation (SO 2 ) is as follows, SO 2 at each pixel can then be derived from PA values at 750 nm and 830 nm as shown in the following equation:

Patients and imaging procedure
The authors prospectively evaluated forty-five patients with breast nodules smaller than 2 cm and with distal margins less than 2 cm from skin surface. All patients received preoperative breast PA/US imaging at our institution between November 2017 and January 2018. Each patient was initially diagnosed with conventional US, X-ray mammography, and/or magnetic resonance imaging by experienced radiologists based on the BI-RADS classification. Grayscale US and CDFI were performed in all 45 patients with BI-RADS classification by 3 radiologists with more than 10-year-experience of breast US diagnosis. Subsequently, PA/US examinations were performed. If there were multiple lesions in a patient, the more suspicious one was selected for PA/US imaging. Excisional biopsy or surgery resection with pathological diagnosis was performed on all the enrolled patients.
Of these forty-five patients, two patients were not imaged successfully due to the system malfunctions. Twenty-six patients had non-intraductal malignant lesions. Of the remaining 17 patients, 9 patients with 10 lesions (1 patient had ipsilateral two lesions) had breast intraductal neoplasm (IDP, n=4; DCIS, n=6), and 8 were either breast fibroma (FA) or breast adenosis. The results for the measurements on 10 breast intraductal lesion and 8 benign lesions (5 fibromas and 3 breast adenosis) are included in this study (Fig. 2). The mean diameter was 1.0 cm for breast intraductal lesions and 1.3 cm for the control group lesions. Detailed information of 17 patients was shown in Supplement 1 Table 1.

Imaging analysis
All images were analyzed and assessed by two physicians with at least 8 years of ultrasound experience. They were blind to the pathological diagnosis and clinical manifestations. Grayscale US and CDFI images were evaluated first, and then PA/US examination results were analyzed one week later. For each lesion, we chose the PA image with strongest PA signals among all acquired images for scoring, and scoring was based only on regions of intra-tumor and regions of peri-tumor within 2 cm depth considering the low SNR in depth larger than 2 cm. In order to further exclude possible noise signals, the tiny and uncertain PA signals were not considered into scoring. If a disagreement occurred, another physician with 13 years of ultrasound experience would review the imaging data, and a consensus was obtained by joint review and discussion.
In CDFI/US and PA/US imaging, vessel scores were first graded based on vessel morphology based on the scoring system according to previous study [18] ( Table 1). The total vessel score was then calculated by adding internal score and peripheral score together. The SO 2 /US fused imaging was also graded based on SO 2 level (score 0, high SO 2 , vessels mostly filled with red; score 1, low SO 2 , vessels mostly filled with blue). At last, BI-RADS assessment was carried out based on scores evaluated from either CDFI/US or PA/US images.

Statistical analysis
Statistical tests (Mann-Whitney U-test) and graphs were calculated using SPSS 21.0 (IBM). P values less than .05 were considered significant.

Difference of vessel scores between PA/US and CDFI/US
Two radiologists assessed the CDFI and PA/US imaging results (n=18) using the scoring system showed in Table 1. Vessel scores for each lesion were shown in Supplement 1 Table 2, numbers of lesions with different scores were summarized in Table 2 and Table 3. The total vessel scores evaluated with PA/US were in general higher than the total vessel scores evaluated with CDFI/US (PA/US total vessel scores: 3.39 ± 1.34, CDFI/US total vessel scores: 1.83 ± 1.62, p = 0.005). CDFI/US and PA/US imaging results of a representative case was shown as below (Fig. 3). For vessel scores inside tumors, there is no significant difference between PAI and CDFI (PA/US: 1.67 ± 1.03, CDFI/US: 1.06 ± 1.21, p=0.110). However, it's worth noticing that with Table 2

. Case numbers with different internal vessel scores in PAI and CDFI a
Internal zone score in PAI (0−3)

Internal zone score in CDFI (0−3)
Total cases  CDFI imaging, 9 cases (9/18, 50%) were scored 0 in internal zone due to no blood signals detected. While with PA/US imaging that found more vessels, only 3 cases were scored 0 (3/18, 16.7%) in the internal zone. For vessel scores in the peripheral zone, PAI scores were significantly higher than CDFI scores (PA/US: 1.72 ± 0.46 CDFI/US 0.78 ± 0.73, p<0.001). With CDFI imaging, 7 cases (7/18, 38.9%) were scored 0 in periphery zone. No cases were scored 0 in the periphery zone with PA/US imaging. Results of vessel scores in both the internal zone and the periphery zone suggested that PA/US imaging is more sensitive to vessels.

Comparison between breast intraductal lesions and control group
In CDFI imaging, there was no significant difference between vessel scores of breast intraductal lesions and control group in both internal zone (breast intraductal lesions: 1.20 ± 1.23, control group: 0.88 ± 1.25, p = 0.503) and peripheral zone (breast intraductal lesions: 0.90 ± 0.88, control group: 0.63 ± 0.52, p = 0.531). With PA/US imaging, significant difference was observed between internal vessel scores of the two groups (breast intraductal lesions: 2.20 ± 0.63, control group: 1.00 ± 1.07, p=0.016). In peripheral zone, there was no significant difference between the two groups (breast intraductal lesions: 1.90 ± 0.32, control group: 1.50 ± 0.53, p = 0.067). CDFI/US and PA/US results of a DCIS case in the right breast of a 56-year-old woman, and a FA case in the right breast of a 31-year-old woman were shown as below. Vessel signals were more abundant in the DCIS case than in the FA case with PA/US imaging as shown in Fig. 4(b), (d). While with CDFI/US, the DCIS case and the FA case showed similar spotty blood signal ( Fig. 4(a), (c)). SO 2 scores of two groups were graded from SO 2 /US fused imaging as shown in Supplement 1 Table 2. Significant difference was observed between breast intraductal lesions and control group (0.90 ± 0.32 vs. 0.25 ± 0.46, p=0.006). Below is the SO 2 /US fused imaging of an IDP case and a FA case (Fig. 5). Most vessel signals showed as blue (low SO 2 ) in the IDP case while in the FA case, green or yellow vessel signals dominated, suggesting a higher SO 2 trend.
To explore the potential diagnostic value of PA/US with the combination of morphological information (vessel score) and functional information (SO 2 score), we defined PA score as the sum of SO 2 score and internal vessel score evaluated from PA/US imaging. Significant  The vessels of FA mostly showed as yellow or red. The SO 2 /US score was graded as 0, demonstrating a higher SO 2 trend. difference was found between breast intraductal lesions and control group (breast intraductal lesions: 3.10 ± 0.57, control group: 1.25 ± 1.04, p = 0.002). The receiver operating characteristic (ROC) curve for differentiating breast intraductal lesions from the control group lesions with PA score was created as shown in Fig. 6. The area under the ROC curve (AUC) was 0.875. With a cut off PA score of 2.5, we got a sensitivity of 90%, a specificity of 87.5%. Fig. 6. ROC curve for differentiating breast intraductal lesions from the control group with PA scores. The AUC was 0.875. With a cut off PA score of 2.5, we got a sensitivity of 90%, and a specificity of 87.5%.
In the control group (n=8), based on CDFI/US, the 8 cases were graded as 3 cases of BIRADS 3, 2 cases of BIRADS 4a, 3 cases of BIRADS 4b, and 1 case of BIRADS 4c, respectively. Based on the analysis of PA/US scoring results, 1 case (1/8, 12.5%) was upgraded (from BI-RADS 4b to BI-RADS 5) and 4 cases (4/8, 50%) were downgraded (from BI-RADS 4a or higher to BI-RADS 4a or 3). These results demonstrated that BI-RADS grading based on PA/US imaging is more coherent with pathology results.

Discussion and conclusion
Our study showed that PA/US functional imaging is more sensitive to vessels than CDFI/US based on total vessel scores and peripheral scores. The total vessel scores evaluated with PA/US was significantly higher than that with CDFI/US (total vessel score of PA/US vs. CDFI/US: 3.39 ± 1.34 vs. 1.83 ± 1.62, p = 0.005). Furthermore, PA/US is especially sensitive to vessels in tumor peripheral zone (peripheral vessel score of PA/US: 1.72 ± 0.46, peripheral vessel score CDFI/US: 0.78 ± 0.73, p<0.001). Though no significant difference between internal vessel scores of two modalities was found at 95% level, there is a trend that more vessel signals could be detected by PA/US imaging (PA/US: 1.67 ± 1.03, CDFI/US: 1.06 ± 1.21, p=0.110). This finding is in consistent with the previous study, where PA/US thyroid imaging revealed more blood vessels than CDFI [27]. The primary reason of this difference existing between PAI and CDFI is the distinct mechanism in these two techniques: CDFI detects blood flow velocity, size and the direction, whereas PAI depends on the optical absorption, which is much less sensitive to these factors. Therefore, PAI can be more sensitive to blood with small flow velocity and blood flow perpendicular to the transducer axial direction than CDFI.
In our statistical analysis of vessel scores of breasts intraductal lesions and non-intraductal benign lesions, the internal vessel scores of breast intraductal lesions evaluated from PA/US was significantly higher than that of control group. In comparison, with CDFI imaging, there was no significant difference between either internal vessel scores or peripheral vessel scores of the two groups. This observation suggests that PA/US imaging has a greater potential in differentiating intraductal lesions from control group due to its higher sensitivity to vessels. In addition to internal vessel scores, we found that SO 2 score could also be used as a distinctive parameter to differentiate breast intraductal lesions from the control group since SO 2 value has a lower trend in breast intraductal lesions than in the control group. By combining PA internal vessel score and SO 2 score, we could reach a favorable AUC of 0.875 with a cut off PA score of 2.5. Similar observation was reported in a previous clinical study with more than 2000 breast tumors imaged by a handheld 2D PA/US functional imaging system. In this study, malignant tumors were found to have a lower trend of SO 2 than benign ones [18], however, no specific comparison was explored between intraductal lesions and benign tumors.
In terms of PA peripheral vessel scores, no significant difference was found between the two groups. However, PA peripheral vessel scores of intraductal breast lesions did have a higher trend than those in the control group. Similar trend has been observed in previous studies. For example, Ganesan's study found increasing periductal flow was a characteristic vascular features of breast intraductal malignancies [28]. In another study, an intact enhancing cyst wall was observed in 61.5% (8/13) of the intracystic breast lesions, especially more in the atypical or malignant ones than benign ones. In this study, Xia postulated the intact enhancing cyst wall may contribute to active angiogenesis and inflammatory infiltration in the tumor peripheral region [29,30]. However, the result of this study was only evaluated and observed in intracystic papillomas with limited case number. Further investigation is still needed to determine whether the peripheral vascular pattern of PA/US imaging is beneficial in diagnosis of breast intraductal lesions.
From our upgrading/downgrading results of PA/US, we found that with PA/US imaging, 40% of the breast intraductal lesions were upgraded and 10% of the cases were downgraded. In the control group, 12.5% of the cases were upgraded and 50% of the cases were downgraded. These results suggested that PA/US imaging could potentially enhance the accuracy of BIRADS grading for high risk breast intraductal lesions and consequently reduce unnecessary biopsy. However, due to the small sample size of this study, further research with larger study sample is demanded.
Our research does have several limitations. First, the breast papillary neoplasms in our study mostly presented as solid mass (9/10) except for one intracystic mass (1/10), while some early stage papilloma which presented as non-solid lesions such as dilated duct with or without associated mass inside were not enrolled. Thus, our findings of PA/US features in evaluation of breast intraductal lesions might be incomplete. Second, the sample size in our study is relatively small, which may affect the statistical analysis power. Third, the semi-quantitative scoring of vessel characteristic is relatively subjective and the interpretation accuracy depends on the experience of physicians, thus may cause some bias in our statistical results. Based on the above, further prospective study with increased sample size will be performed to investigate the reliability of our results.
Our results suggested that besides providing tissue functional information, PA/US imaging is also more sensitive to display the morphological features of vessels than CDFI/US imaging in breast intraductal lesions. Meanwhile, with the combination of internal vessel scores and SO 2 scores evaluated from PA/US imaging, a favorable sensitivity and specificity could be achieved to differentiate breast intraductal lesions from typical benign tumors.