Noninvasive assessment of endometrial fibrosis in patients with intravoxel incoherent motion MR imaging

Recently, few noninvasive methods have been reported to evaluate endometrial fibrosis. Our study was to investigate the feasibility of intravoxel incoherent motion (IVIM) MR imaging in the detection of endometrial fibrosis in patients with intrauterine injury. 30 patients with hysteroscopy-confirmed endometrial fibrosis and 28 healthy women were enrolled to undergo MR examination including the IVIM sequence. Endometrial thickness (ET); apparent diffusion coefficient (ADC); and IVIM parameters, including pure diffusion coefficient (D), pseudodiffusion coefficient (D*) and vascular fraction (f) were evaluated. A multivariable model combing ADC, D, and f values using binary logistic regression analysis was built to diagnose endometrial fibrosis. Endometrial fibrosis patients demonstrated lower endometrial ADC, D, f values and ET (all p < 0.05). The multivariable model, ADC, D, f values and ET performed well in diagnosing endometrial fibrosis with AUC of 0.979, 0.965, 0.920, 0.901 and 0.833, respectively. The multivariable model revealed a better diagnostic accuracy than D, f and ET (all p < 0.05). Although ADC achieved a better diagnostic value than ET (z = 2.082, p < 0.05), no difference in AUC was shown among ADC, D, and f (all p > 0.05); between ET and D (p > 0.05); and between ET and f (p > 0.05). The reproducibility of ADC, D, f and D* values in patients with endometrial fibrosis and healthy women were good to excellent (ICC: 0.614–0.951). IVIM parameters exhibit promising potential to serve as imaging biomarkers in the noninvasive assessment of endometrial fibrosis.


Scientific Reports
| (2021) 11:12887 | https://doi.org/10.1038/s41598-021-92383-w www.nature.com/scientificreports/ As a quantitative imaging technique, intravoxel incoherent motion (IVIM) MR imaging can be used to noninvasively investigate diffusivity and microcapillary perfusion in biological tissues 9,10 . In recent years, it has been successfully used to stage liver fibrosis, to evaluate the degree of fibrosis of the kidney and pancreas, and to detect parotid gland and intestinal fibrosis [11][12][13][14][15] . Based on those studies and the histopathology of endometrial fibrosis, it is presumable that IVIM MR imaging might serve as a potential imaging biomarker for evaluating endometrial fibrosis.
The purpose of this study was to investigate the difference in IVIM-derived parameters of the endometrium between patients with endometrial fibrosis and healthy women and to explore the diagnostic performance of IVIM parameters in endometrial fibrosis.

Results
Study population. All the patients underwent dilation and curettage (D&C) once or several times, and some of them experienced transcervical resection of adhesions. The clinical demographics of patients and healthy women are detailed in Table 1. The process of patient inclusion and exclusion is displayed in a flow chart (Fig. 1). Four patients and one healthy woman were excluded for poor image quality caused by massive uterine effusion or bleeding of the uterine cavity. Finally, 30 patients (mean age: 33.30 years; range 27-42 years) and 28 healthy women (mean age: 29.00 years; range 24-38 years) were included in this study. Table 2, the mean ADC, D, f values and ET was significantly lower in patients with endometrial fibrosis (all p < 0.05). However, no significant difference in the D* value was noted between patients and healthy women (p > 0.05). Figures 2 and 3 show T2WI, DWI and the corresponding parametric maps (ADC, D, f and D* maps) of the uterus in a healthy woman and a patient. Table 3, the performance of the multivariable model, ADC, D, f values and ET in diagnosing endometrial fibrosis were all excellent with AUCs of 0.979, 0.965, 0.920, 0.901, and 0.833, respectively, and the multivariable model had the highest AUC. Figure 4 shows the ROC curves of the multivariable model, ADC, D, f values and ET for distinguishing patients with endometrial fibrosis from healthy women. The multivariable model revealed a better diagnostic performance than D, f and ET (z = 1.980, 2.190, 2.406, respectively, all p < 0.05). ADC achieved a better diagnostic value than ET (z = 2.082, p < 0.05). No difference in AUC was shown between the multivariable model and ADC (p > 0.05), among ADC, D, and f (all p > 0.05); between ET and D (p > 0.05); and between ET and f (p > 0.05). The false negative rates for ADC, D and f values in the diagnosis of endometrial fibrosis were 20%, 20%, and 25%, respectively. The false positive rates for ADC, D and f values in the diagnosis of endometrial fibrosis were 0%, 14.3%, and 6.7%, respectively. Table 4, the intra-and interobserver agreements in the measurements of ADC and D values in patients with endometrial fibrosis and healthy women were excellent (all ICC > 0.800); for the measurements of f and D* values, the intra-and interobserver agreements were good to excellent (ICC: 0.614-0.935).

Discussion
Our pilot study demonstrated differences in IVIM parameter values and ET between patients with endometrial fibrosis and healthy women. In addition, endometrial ADC, D, f values, ET and the multivariable model had good efficiency for diagnosing endometrial fibrosis. The intraobserver and interobserver agreements of endometrial ADC, D and f value measurements were good.
In the present study, we observed that ADCs and diffusion-linked component D values decreased significantly in the fibrous endometrium. This finding could be explained by the fact that normal endometrial glands and stromal cells could become atrophic and replaced by simple cuboidal epithelium incrementally as the fibrosis of endometrium progressed 1,16 . The subsequent denser cellularity and abundant accumulation of extracellular matrix (ECM), especially fibrillary collagens, could inhibit the random motion of water molecules, which might result in the reduction of ADC and D values 3,10,16 . Similar to other organ fibrosis, due to the presence of endometrium perfusion rather than pure diffusion restrictions, ADC values were greater than the corresponding D values 10,15,17 . Thus, compared with the biexponential model, a monoexponential model could overestimate the water diffusion in the fibrotic tissue.
According to previous studies, it is widely held that blood perfusion decreases in fibrotic tissue due to concomitant alterations in tissue microcirculation, including damage to capillary networks, proliferation of connective tissue and increased resistance to blood flow 10,15 . This might explain the lower f value in fibrotic    www.nature.com/scientificreports/ endometrium compared with normal endometrium in this study. The decreased blood perfusion might lead to embryo implantation dysfunction and consequent infertility or spontaneous abortion 1,3 . Thus, we hypothesize that the f value may have the potential to serve as an imaging biomarker reflecting endometrium perfusion without the administration of contrast agents. Interestingly, in this study, the perfusion-related parameter D* value exhibited no difference between patients and healthy women. Generally, the D* value could reflect endovascular blood flow velocity within tissue, which was significantly lower in fibrotic organs 16,18,19 . However, D* value is well known for its large standard deviation,   www.nature.com/scientificreports/ data instability and its dependence on signal-to-noise ratio (SNR) 14,19,20 , which limits its clinical applicability. Further studies are warranted to focus on these deficiencies and improve the veracity of D*. Furthermore, it is worth noting that all of the participants in this study underwent MR examination during the periovulation phase due to the existence of the physiological dynamic changes in endometrial thickness and microstructure over the different phases of the menstrual cycle 21 .
Previous studies confirmed that trauma to the basal layer of the endometrium caused endometrial fibrosis and affected the regeneration of endometrial epithelial cells, which led to a thin endometrium 1 . Although thin endometrium could be used as one of the predictive markers of endometrial fibrosis, it does not mean that all thin endometriums are from patients with endometrial fibrosis. In other words, among women with thin endometrium, only those whose endometrium has abnormal microstructure and function are diagnosed with endometrial fibrosis. It has been reported that some women with endometrium thinner than 4 or 5 mm may still have normal endometrial function and become pregnant successfully 22,23 . Thus, in this study, IVIM-DWI as a functional imaging technique demonstrated superior capability to conventional MR imaging according to reflecting the biologic abnormality of the body at a cellular level 9,10 .
This study initially revealed the feasibility of IVIM MR imaging for the evaluation of patients with endometrial fibrosis. Significantly lower ADC, D and f values were detected in the endometrium of patients with endometrial fibrosis. Though ADC showed higher AUC compared to D and f, no significant differences were found. Moreover, the IVIM parameter f could reflect the perfusion changes in endometrial fibrosis 9 , which cannot be obtained with the conventional DWI model. So we insisted that compared with DWI, IVIM parameters could provide added value on the evaluation of the perfusion changes in fibrotic endometrium. The multivariable model showed the highest diagnostic performance in endometrial fibrosis, which indicated that the multivariable model might provide better diagnostic performance than the single IVIM parameter. There were relative low false negative/ positives for ADC, D and f values in the diagnosis of endometrial fibrosis. Not all methods fail on the same patient, and the IVIM parameters are complementary in diagnosing endometrial fibrosis. In addition, ADC, D and f values exhibited good intraobserver and interobserver agreements. In a word, this study revealed that both the multivariable model and IVIM parameters had promising potential in the diagnosis of endometrial fibrosis, which might be helpful for clinicians to implement antifibrotic therapy, and to conduct dynamic follow-ups.
There are several limitations in this study. First, the sample size was relatively small, and patients with early endometrial fibrosis were absent. Nevertheless, the feasibility of utilizing quantitative MR imaging, including IVIM parameters and ET, to evaluate endometrial fibrosis was demonstrated. Second, due to the small sample size, we were unable to further stratify grades of endometrial fibrosis according to the IVIM parameters. Therefore, in future studies, a large cohort of patients will be recruited. Third, as the appropriate number of b values suitable for endometrial fibrosis remains unknown, further investigation is required.
In conclusion, our study confirmed the significant differences in intravoxel incoherent motion (IVIM) derived parameters of the endometrium between patients with endometrial fibrosis and healthy women. IVIM parameters provided functional features of the fibrotic endometrium, in which ADC, D and f values performed well in differentiating fibrotic endometrium from normal endometrium. It is conceivable that IVIM MR imaging has the potential to serve as an imaging biomarker in the noninvasive diagnosis of endometrial fibrosis.

Materials and methods
Subjects. This prospective study was approved by the ethics committee of the Institutional Review Board of Nanjing Drum Tower Hospital, and written informed consent was obtained from all participants. All experiments were performed in accordance with relevant guidelines and regulations. From October 2018 to December 2019, 34 patients with endometrial fibrosis confirmed by hysteroscopy (mean age: 33.56 years; range 27-42 years) and 29 healthy women with normal endometrium who served as the controls (mean age: 29 years; range 24-38 years) were recruited consecutively in our study.
The inclusion criteria for patients were as follows: (1) clinically diagnosed as infertile women with a history of intrauterine surgery, such as D&C or transcervical resection of adhesion; (2) endometrial scars confirmed by hysteroscopy; (3) no history of other severe uterine diseases, including adenomyosis, large intramural myomas, endometrial tuberculosis and severe congenital uterine malformations; (4) normal ovarian function; and (5) no MRI contraindications, such as cardiac pacemakers, cochlear implants and claustrophobia.
The criteria for inclusion in the healthy women group were as follows: (1) women of reproductive age with regular menstrual cycle and normal menstrual volume; (2)  www.nature.com/scientificreports/ view = 160 mm × 160 mm, slice thickness = 3 mm, intersection gap = 0.5 mm, and NSA = 2). The scanning time of IVIM was approximately 2 min 37 s, and the total scanning time was approximately 8 min 51 s. All participants underwent the examinations successfully without any discomfort or side effects.
MR analysis. MR images were independently reviewed and analyzed by two radiologists (Li Zhu and Zhengyang Zhou, with 6 and 15 years of experience in gynecology radiology, respectively) who were blinded to the clinical information of the patients. The final results of one participant were calculated from averaged values of the two radiologists. IVIM data were repeatedly measured by the first radiologist one month later for intra-and interobserver reproducibility analyses. All MR images were transferred to the Picture Achieving and Communication System (PACS) Workstation. Endometrial thickness (ET) was measured on midsagittal T2-weighted images.
The IVIM data were evaluated using DWI-Tool developed by Philips in IDL 6.3 (ITT Visual Information Solutions, Boulder, CO, USA), and the D, D*, and f maps were generated automatically. The quantitative values of D, D* and f were calculated using the biexponential model raised by Le Bihan 9 with the following equation: in which Sb and S0 denote the mean signal intensity at a specific b value and when the b value is 0 s/mm 2 , respectively. The ADC maps and the corresponding values were estimated by using the monoexponential fit function at all 9 b values: The specific slice of DWI with the largest endometrial section was selected referring to the corresponding sagittal T2-weighted images. Then, a region of interest (ROI) was manually drawn to include as much endometrium as possible. The ROI was positioned carefully inside the contour of the endometrium in the upper 2/3 uterus corpus and delineated on a relatively homogeneous endometrial region to avoid the areas with large degree of heterogeneity in the parametric maps. The ROIs were automatically propagated between the monoand biexponential models to produce the corresponding D, D*, f and ADC values. Three ROIs were drawn repeatedly by each radiologist, and the average value of a total of six ROIs served as the representative result for subsequent statistical analysis. Statistical analysis. Statistical analysis was performed with SPSS 22.0 (SPSS Inc., IL, USA). Continuous variables with a normal distribution are presented as the means ± standard deviations (SD), whereas nonnormally distributed variables are presented as the median or interquartile range. Independent t tests were used to compare the differences in ET, ADC and D values between patients and healthy women. For the comparison of D* and f values obtained from patients and healthy women, the Mann-Whitney U test was used. A multivariable model combing ADC, D, and f values using the binary logistic regression analysis was built to diagnose endometrial fibrosis. Receiver operating characteristic (ROC) curve analysis was performed to evaluate the diagnostic performance of ADC, D, f, ET and the multivariable model for the discrimination of endometrial fibrosis from normal endometrium using the DeLong test in MedCalc 19.1.0.0 (MedCalc statistical software, Mariakerke, Belgium). The area under the curve (AUC) and the optimal cutoff value of each parameter for achieving the best diagnostic accuracy were calculated using the De Long test, and an AUC greater than 0.80 defined as excellent diagnostic efficacy. Additionally, intra-and interobserver reproducibility were evaluated using the intraclass correlation coefficient (ICC), which was classified as excellent ( 0.81-1.00), good ( 0.61-0.80), moderate (0.41-0.60), fair (0.21-0.40), and poor (0.00-0.20) 24 . p-values less than 0.05 were considered statistically significant.

Data availability
The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.