Tumor lesion detection in patients with cervical cancer by indocyanine green near-infrared imaging

To investigate the feasibility and accuracy of near-infrared fluorescence (NIRF) imaging for detecting the extent of tumor invasion in cervical cancer using indocyanine green (ICG). We enrolled 51 patients who were diagnosed with cervical cancer with FIGO stage IB1-IIA2 disease. Patients were administered indocyanine green (ICG) at a dose of 5 mg/kg 24 h prior to surgery. A customized near-infrared fluorescence (NIRF) imaging system was used to identify the extent of tumor invasion when radical hysterectomy specimens were harvested. The relationship between tumor fluorescence intensity and clinicopathological characteristics was analyzed. Of the 51 enrolled patients, 3 patients did not have residual tumors after cervical conization, and tumor lesions were identified by NIRF imaging in all the remaining 48 patients. The results of NIRF imaging were in agreement with the postoperative pathological findings in 95.8% of the patients with stromal invasion, 100% of those with surgical margin invasion, 100% of those with parametrial tumor involvement, and 100% of patients with uterine corpus invasion. The mean signal-to-background ratio (SBR) of the cervical tumors was 2.91 ± 1.64, and the SBR was independent of clinicopathological characteristics. Fluorescence microscopy confirmed that ICG fluorescence was present in the tumor nests. NIRF imaging enables objective, accurate, and safe identification of tumor invasion during cervical cancer surgery. ClinicalTrials.gov NCT04224467.


Introduction
Cervical cancer is the fourth most common cancer after breast cancer, rectal cancer, and lung cancer in women worldwide [1], and it seriously threatens women's health, especially those who live in low-resource regions or socioeconomically weaker areas of society [2].
The degree of tumor invasion determines the primary treatment options for patients with cervical cancer. Due to the advantages of magnetic resonance imaging (MRI), such as the high resolution of soft tissue and multiplanar, multisequence, and multiparameter imaging, MRI is often used to evaluate the preoperative tumor invasion range in patients with cervical cancer, beneficially affecting the choice of primary treatment [2,3]. However, MRI examination cannot intraoperatively evaluate the extent of tumor invasion and cannot determine whether the tumor has invaded the surgical margin of the vagina [4]. In the cases of positive surgical margins, some patients require additional postoperative radiotherapy, which increases the economic burden and complication risk. Radical trachelectomy (RT) is an important fertility-sparing treatment for cervical cancer patients, but some patients require secondary surgery or postoperative radiotherapy due to positive surgical margins [5,6]. Postoperative adjuvant radiotherapy may lead to ovarian or endometrial damage and failure of fertility preservation [7]. Therefore, intraoperative evaluation of tumor invasion, especially whether the surgical margin is involved, has important clinical significance.
Intraoperative frozen sections can be used to assess vaginal margins, but they prolong the operation and anesthesia time as well increase the risk; in addition, false positives and false negatives can occur with frozen pathology [8,9]. Intraoperative gross examination of the extension of tumor invasion depends heavily on the experience of the surgeon, lacks objective criteria, and has low specificity and sensitivity [9]. The development of a new method to intraoperatively estimate tumor invasion during cervical cancer surgeries is warranted.
Near-infrared f luorescence (NIRF) imaging has become one of the most attractive intraoperative navigation techniques for localizing tumor tissue and identifying tumor margins. Indocyanine green (ICG), a safe imaging dye that produces few allergic reactions [10], has been the most widely used NIRF dye to date for the detection of several tumors, such as pulmonary nodules [11,12], intracranial tumors [13,14], mediastinal tumors [15], and pancreatic tumors [16]. Previous studies have indicated that ICG remains in tumors for a prolonged time after intravenous injection due to the enhanced permeability and retention (EPR) effect [11][12][13][14][15][16]. In gynecological surgeries, NIRF imaging has recently been proposed as an intraoperative tool for the detection of sentinel lymph nodes or abdominal pelvic lymph nodes in both clinical trials [17,18] and case reports [19]. However, to the best of our knowledge, there are no published clinical studies on NIRF imaging using ICG for the extent of cervical tumor invasion.
Based on clinical data on NIRF imaging of other solid tumors and our preliminary experiments, we hypothesized that the extension of tumor invasion in cervical cancer can be intraoperatively identified by intravenous injection of ICG. In this pilot study, we aimed to explore whether NIRF imaging can be used to effectively and safely detect the extent of tumor invasion in cervical cancer as a novel imaging strategy and whether the tumor fluorescence intensity is independent of clinicopathological characteristics.

Study design
This study was approved by the Ethics Committee of Nanfang Hospital, Southern Medical University (Ethics no. NFEC-2019-242), and all enrolled patients provided informed consent. The identifier of the clinical trial is NCT04224467. The inclusion criteria were as follows: patients with cervical cancer; preoperative FIGO stage IB1-IIA2 (FIGO 2009 staging system); age > 18 years; underwent open radical hysterectomy plus pelvic lymphadenectomy; preoperatively normal liver function; and no ICG or iodine allergies. The exclusion criteria were as follows: thyroid pathology; a history of pelvic surgery, radiotherapy or chemotherapy; current pregnancy; and considered unsuitable for enrolment by the clinicians. Finally, 51 patients with cervical cancer were enrolled in this study between March 2020 and May 2021.

Fluorescence imaging system and ICG administration
All subjects were imaged using a customized NIRF device (DPM-I, Beijing Digital Precision Medicine Co., Ltd.), which was mainly developed based on our previous study. The device can realize simultaneous color (380-665 nm) and fluorescence (840-900 nm) imaging and registration [20][21][22].
We purchased ICG from Dandong Yichuang Pharmaceutical Co., Ltd., China. An intravenous injection of ICG at a 5-mg/ kg dose was administered to patients 24 h preoperatively. The injection time and dose of ICG were determined based on the protocols used for fluorescence imaging of other solid tumors [11][12][13][14][15][16]20]. To detect allergic reactions in a timely manner, patients were monitored for 15 min after ICG injection.

Imaging procedure
During the operation, the surgeons performed a conventional open radical hysterectomy. Once the radical hysterectomy specimens were harvested, they were imaged ex vivo in a lightproof box in the operating room before pathological analysis.
The NIRF imaging method was used to identify the extent of tumor invasion when radical hysterectomy specimens were harvested. First, the specimens were dissected from the anterior wall of the uterus and vagina in a Y-shape, and cervical tumors were visualized with an NIRF imaging device. Then, the depth of stromal invasion (superficial stromal invasion was defined as a depth of stromal invasion < 1/2 of the stromal depth, and deep stromal invasion was defined as a depth of stromal invasion ≥ 1/2 of the stromal depth), parametrial tumor involvement, surgical vaginal margin invasion, and uterine corpus invasion were also assessed using the NIFR imaging method. The final histopathological diagnosis was determined by a single pathologist at our institution who was blinded to the study, and this analysis served as the gold standard for the extent of tumor invasion. In addition, ICGrelated adverse events and the time spent for intraoperative NIRF imaging were recorded.

Fluorescence microscopy
Frozen sections were consecutively obtained at 10-μm intervals. For two consecutive sections, one was stained with hematoxylin-eosin (H&E) to observe the location of the cancer nest, and the other was used to observe the fluorescent signal after labeling nuclei with 4',6-diamidino-2-phenylindole (DAPI) because H&E staining may compromise ICG fluorescence signal observation. The sections were then examined using an LSM980 laser confocal microscope (Zeiss, Germany).

Statistical analysis
The signal-to-background ratio (SBR) of cervical tumors was quantitatively analyzed based on the recorded images. First, we used ImageJ (Image processing Software, National Institutes of Health, USA) to delineate a region of interest (ROI) in the target cervical region and calculate the mean grey value of the ROI (V signal ) within a range of 0-255. The background brightness (V Background ) was measured by calculating the mean grey value of the surrounding tissues. The SBR was then calculated by dividing the V signal by the V Background . Each patient was evaluated three times to reduce errors.
All statistical procedures were processed with SPSS 23.0 statistical software (Statistical Product and Service Solutions, Chicago, USA). Categorical variables are expressed as whole numbers and percentages, and they were compared using the chi-square test. Continuous variables are expressed as the mean values ± standard deviation. The sensitivity and specificity were also computed. The independent sample t test or one-way ANOVA was used to compare the differences in the SBR among different groups. Multiple linear regression was used to estimate the effect of clinicopathological characteristics on the SBR, including age, FIGO stage, histology, differentiation, tumor size, stromal invasion, lymphovascular space invasion (LVSI), parametrial tumor involvement, and lymph node metastasis. A P value of 0.05 or lower was considered statistically significant.

Clinical data
A total of 51 patients were included in this study, and their mean age was 49.35 ± 10.61 years. The clinicopathological characteristics of the patients are shown in Table 1. Four patients with preoperative FIGO IB1 stage, 4 patients with preoperative FIGO IB2 stage, 2 patients with preoperative FIGO IIA1 stage, and 1 patient with preoperative FIGO IIA2 stage were corrected to IIIC1p stage due to lymph node metastasis, and 1 patient with IIA2 stage was corrected to IIB stage due to parametrial tumor involvement. Among these patients, 9 patients underwent cervical conization, and 3 of them had no residual tumor confirmed by final pathology. No ICG-related adverse events occurred in any patient. The mean time spent in the NIRF imaging procedure was 4.85 ± 1.54 min.

Detection of cervical cancer by NIRF imaging
Three patients had no residual tumor in the resected uterine specimen after cervical conization. Tumor tissues were identified using the NIRF imaging method in specimens from the remaining 48 patients, and the mean SBR of cervical carcinoma was 2.91 ± 1.64. Figure 1 shows that cervical tumor lesions were visualized by NIRF imaging regardless of whether the tumor had deep or superficial stromal invasion. Figure Table 2 shows the results of stromal invasion analysis with the use of NIRF imaging and postoperative pathology. Of the 48 cases with fluorescent tumors, NIRF imaging and final pathology were in agreement in 46 patients. Two patients had false negatives; they were diagnosed with superficial stromal invasion by NIRF imaging but were advanced to deep stromal invasion according to the final pathological analysis. All deep stromal invasion identified by the NIRF method was correlated with the final pathological analysis. As a result, the sensitivity and specificity values of NIRF imaging in deep stromal invasion were 93.5% and 100%, respectively, with an accuracy of 95.8%. Table 3 shows the results of surgical margin invasion analysis with the use of NIRF imaging and postoperative pathology. Of the 48 patients, 46 were found to have no surgical margin invasion by the former method, and the final vaginal margin was pathologically confirmed to be without tumor invasion. The other two patients were determined to have fluorescent signals at the primary surgical margin using NIRF imaging, and their vaginas were resected until no fluorescent signal was observed at the final vaginal margin. Pathological analysis showed that the two primary vaginal margins with fluorescence signals were representative of tumor invasion, and all the final vaginal margins were demonstrated to be free of tumor invasion. In these two patients, one patient still received postoperative radiotherapy due to lymph node metastasis, and the other patient had only one medium-risk factor (tumor size > 4 cm) and did not receive postoperative radiotherapy. Table 4 shows the results of the parametrial tumor involvement analysis, and Table 5 shows the results of the uterine corpus invasion analysis with the use of NIRF imaging and postoperative pathology. The concordance between NIRF imaging and postoperative pathology was 100%.

Fluorescence signal intensity is independent of clinicopathological characteristics
For the 48 fluorescent cervical tumors, we hypothesized that the fluorescence signal intensity of the tumors may be dependent on the clinicopathological characteristics of the patients. We compared the SBR of patients with different clinicopathological characteristics, and we found that the mean SBR of patients with different preoperative FIGO stages, differentiation, histology, stromal invasion, tumor size, LVSI, parametrial involvement, and lymph node metastasis was not statistically significant (all p > 0.05, Fig. 3). Considering the possible interaction among the above variables, which may have affected the results of the univariate analysis, we conducted a multivariable regression analysis to predict the SBR. The results indicated that preoperative FIGO stage (IB1, IB2, IB3, IIA1, or IIA2), age (years), histology (squamous cell carcinoma, adenocarcinoma, or adenosquamous cell carcinoma), differentiation (well, moderate, or poor), tumor size (measured by columns show the preoperative sagittal T2-weighted MR images, color images, NIRF images, and merged images, respectively preoperative MRI; ≤ 2 cm, > 2 cm and ≤ 4 cm, or > 4 cm), stromal invasion (superficial or deep), LVSI (yes or no), surgical margin invasion (yes or no), parametrial tumor involvement (yes or no), and lymph node metastasis (yes or no) did not predict the SBR value in the regression model.

Fluorescent microscopy
Fluorescence microscopy confirmed that ICG accumulated in the tumor tissue regardless of whether the tumor had deep or superficial stromal invasion. Figure 4A-D show that the ICG fluorescent signal was detected in tumors with deep   stromal invasion, and Fig. 4E-H show that the ICG fluorescent signal was also detected in tumors with superficial stromal invasion. Thus, fluorescence signals were detected in both cancer nests and tumor tissues in lymphatic vascular spaces, while fluorescence signals were rarely detected in the surrounding normal cervical tissues.

Discussion
The aim of the present study was to determine whether NIRF imaging can effectively identify the extent of tumor invasion in cervical cancer using ICG. In this pilot study, all 48 tumor lesions were visualized by fluorescence after intravenous injection of 5 mg/kg ICG 24 h before surgery. The mean SBR of the cervical lesions was 2.91 ± 1.64, which indicated satisfactory imaging results. Fluorescence microscopy also confirmed that the ICG fluorescence signal was detected in tumor tissues. The specificity, sensitivity, and accuracy of NIRF imaging were high for evaluating surgical margin invasion, parametrial tumor involvement, and uterine corpus invasion. Two patients with primary positive vaginal margins were detected by the NIRF imaging technology as follows: one patient avoided adjuvant radiotherapy after resection of the vagina; and the other patient received postoperative radiotherapy due to lymph node metastasis. In the evaluation of stromal invasion, NIRF imaging indicated two cases of deep invasion as false negatives. Further analysis of the pathological sections of these two cases showed that the tumor infiltrated into the deep cervical stroma like a tree root. In most cases, however, the pattern of cervical tumor interstitial invasion was whole tumor invasion. These false negatives indicated that NIRF imaging technology failed to identify patients with deep invasion similar to a tree root, indicating that NIRF imaging technology needs to be improved to solve this problem.
In the present study, the intraoperative procedure of NIRF imaging did not increase the additional complications of the enrolled patients. Although the NIRF imaging procedure required approximately 5 min, the surgical team also performed other operations at the same time, such as hemostasis and suturing vaginal margins. According to our observations, the overall operation time was not significantly prolonged. This is the first study to demonstrate the role of NIRF imaging for the identification of tumor lesions in cervical cancer. Typically, clinicians determine the extent of tumor invasion by observation and palpation during surgery, which is highly subjective and experience-dependent [23]. Intraoperative frozen pathology is not usually used during surgery due to the increased anesthesia and operative time. In some cases, postoperative pathology ends up being confirmed as positive for vaginal or parametrial surgical margins, increasing the rate of postoperative radiation therapy or recurrence. NIRF imaging provides a feasible method for detecting margin invasion during cervical cancer surgery. If a positive fluorescent margin is detected, the surgical margin will be re-excised prior to completion of surgery. In this study, two patients with positive primary surgical margins underwent re-excision of the vaginal margin until no ICG fluorescence signal was observed. One patient was spared postoperative radiotherapy due to only one intermediate risk factor in the final pathology report, and the other patient still received postoperative radiotherapy due to lymph node metastasis. If there is no fluorescent signal at the surgical margin, risk factors for recurrence or death from cervical cancer, including cervical stromal invasion, parametrial tumor involvement, and uterine corpus invasion, will be analyzed by NIRF imaging. Although intraoperative identification of these factors does not appear to affect the postoperative management of patients, these risk factors are strongly associated with para-aortic lymph node metastasis [24]. Once these factors are detected by intraoperative NIRF imaging, the clinician will be alerted to carefully explore whether the para-aortic lymph nodes are enlarged. Para-aortic lymph node dissection or sampling will be performed when the para-aortic lymph nodes are enlarged or when lymph node metastasis is suspected.
Another potential application of this new technology is to identify the surgical margins of the uterus during RT surgery, thus avoiding secondary surgery or postoperative radiotherapy due to positive surgical margins. In future studies, we will explore the surgical success and fertility-sparing rates of RT guided by NIRF imaging.
We also analyzed the correlation between clinicopathologic characteristics and the SBR value of tumor lesions by univariate and multivariate analyses. We found that the SBR value was independent of clinicopathological characteristics, including age, FIGO stage, histology, differentiation, tumor size, stromal invasion, LVSI, parametrial tumor involvement, and lymph node metastasis. The SBR is an index used to evaluate the NIRF imaging quality. The SBR quantifies the significance of the cervical tumor signal and intuitively reflects the ratio of the average grey value of the target signal and the background. The mean SBR of cervical tumors was 2.91 ± 1.64, showing satisfactory NIRF imaging results, and the SBR value did not affect the postoperative management of patients. Because the SBR value was independent of tumor size, NIRF imaging technology may be suitable for small cervical tumor imaging. In the future, we will design fluorescent colposcopy devices to realize accurate and fluorescent biopsy of cervical tumors.
ICG is the only near-infrared fluorescent dye approved by the FDA and the Chinese Food and Drug Administration (CFDA) for clinical use in humans [10]. After intravenous injection, ICG binds rapidly to plasma protein in the blood and is metabolized rapidly by the liver, resulting in a short half-life; thus, ICG is a safe water-soluble dye. ICG has satisfactory optical properties due to its particular excitation wavelength and emission wavelength. ICG NIRF imaging techniques have yielded favorable imaging results for the identification of solid tumors [11][12][13][14][15][16], such as lung tumors, meningeal tumors, and mediastinal tumors. Although fluorescence visualization of cervical tumors is not an indication for ICG, the Ethics Committee of Nanfang Hospital, Southern Medical University, approved this study in consideration of the safety of ICG and its application in other solid tumors.
The mechanism of intraoperative visualization of solid tumors may be derived from the EPR effect. The EPR effect is a property in the tumor microenvironment in which some substances, such as ICG plasma protein conjugates, passively accumulate in the tumor stroma due to the presence of defective endothelial cells and wide fenestrations in nascent blood vessels [25,26]. After 24 h, ICG in normal tissue is metabolized, but ICG is still present in tumor tissue at high concentrations. In the present study, ICG fluorescence signals were detected in cancer nests but were rarely detected in the surrounding normal cervical tissues. Therefore, the mechanism of intraoperative NIRF imaging of cervical tumor lesions may also be explained by the EPR effect.
This pilot study also demonstrated that NIFR imaging had shortcomings in identifying the extent of tumor invasion in cervical cancer. Because the maximum depth of penetration of near infrared light is approximately 1 cm and the thickness of the normal cervix is approximately 2 cm, it is difficult to determine the extent of tumor invasion before the uterus is dissected. Thus, it is challenging to visualize cervical tumors during surgical procedures. In the future, we will further improve this application by using the more penetrating near-infrared II window.
The present study had several limitations. First, this was a single-center pilot study, and the positive predictive value of NIRF imaging in detecting the extent of tumor invasion was affected by the small number of positive cases. Therefore, more patients from multiple centers need to be recruited to verify the ability of NIRF imaging for tumor lesion detection, and a much larger number of positive cases should be investigated. Second, the SBR is an arbitrary number and may not be a true reflection of tissue fluorescence. Third, the present study did not include cervical cancer patients with uncommon pathological types, and this technique may not be suitable for various pathological types of cervical cancer. Fourth, ICG is a nontargeted contrast agent, and a better imaging effect may be obtained by novel targeted contrast agents, such as folate receptor alpha-targeted NIR contrast agents [27].

Conclusions
This is the first-in-human demonstration of identifying tumor lesions during cervical cancer surgery with NIRF imaging, which enables objective, accurate, and safe identification of tumor invasion. However, these preliminary results need to be confirmed by a larger multicenter study.