Normalized apparent diffusion coefficient: a novel paradigm for characterization of endometrial and subendometrial lesions

Abstract

Objectives:

To assess the role of normalized apparent diffusion coefficient (ADC) in characterization of endometrial and subendometrial masses, measured as a ratio of the mean ADC of the pathology to mean ADC of two different internal controls, normal myometrium and gluteus maximus muscle, referred to as nADC_m and nADC_g respectively.

Methods:

55 females with pathologically proven endometrial and subendometrial lesions, including 27 cases of endometrial carcinoma, and 28 cases of benign masses were enrolled in this prospective study and assessed with single-shot echoplanar diffusion-weighted imaging. The normalized and absolute ADC of the lesions, measured by two radiologists, were compared in different pathologies and receiver operating characteristics (ROC) performed to distinguish benign and malignant endometrial masses. In the endometrial carcinoma group, the ADC values were further compared with tumor grade and subtype.

Results:

There was good interobserver agreement (>0.800) for both internal controls, however it was higher for myometrium [intraclass correlation coefficient-0.92; confidence interval (0.86–0.95)] than gluteus maximus muscle [ICC-0.84; CI (0.72–0.90)]. There were statistically significant differences in absolute ADC (p-0.02), nADC_m (p-0.02) and nADC_g (p < 0.0001) of benign and malignant endometrial masses.

Conclusion:

Normalized ADC is useful to distinguish benign and malignant masses with comparable accuracy as absolute ADC.

Advances in knowledge:

Normalized ADC represents an easily measurable quantitative parameter which limits the influence of endogenous and exogenous factors that affect its reproducibility.

Introduction

Diffusion-weighted imaging (DWI) is a functional imaging technique that depicts tissue characteristics based on the microscopic random diffusion of water molecules. Apparent diffusion coefficient (ADC) values represent a derived quantitative index describing this water movement which may be limited by factors such as interactions with cell membranes and macromolecules.¹ In general, malignant lesions show a lower ADC than benign lesions, which can be explained by a combination of higher cellularity and in-vivo tissue disorganization.^2–4

The utility of ADC values in characterization of endometrial and subendometrial masses has been described in previous studies, including distinguishing benign and malignant endometrial masses and correlation of ADC with prognostic parameters of endometrial carcinoma.^5–12 However, there is no consensus regarding the results and the optimal technique for DWI, which can be due to differences in the acquisition and post-processing parameters in different studies. We propose that normalization of ADC by comparison with an internal control may help reduce this variability and allow better comparison across different studies. The preferred control should be one which will show constant pathological nature across different patients and preferably is in the vicinity of the target lesion, such that measurements may be obtained in the same plane as the index lesion. In this prospective study, we measured the normalized ADC of endometrial masses, which is the ratio of the mean ADC of the endometrial mass to mean ADC of two different controls, normal myometrium and gluteus maximus muscle as internal references, referred to as nADC_m and nADC_g respectively. The purpose of this study is to examine the role of absolute and normalized ADC in characterization of endometrial and subendometrial masses, and to assess its utility in predicting prognostic parameters of endometrioid carcinoma such as histological grade and subtype.

Methods and materials

Study population

This prospective study performed at a single tertiary medical center over a period of 36 months from July 2016 to July 2018 was approved by the institutional ethics committee. Written informed consent was obtained from all study participants. 120 females with abnormal uterine bleeding suspected to have focal or diffuse endometrial abnormality on preliminary ultrasound were included in this study. Pregnant females, females with contraindications to MR and patients without histological confirmation were excluded. A total of 55 females with pathologically proven endometrial and subendometrial lesions who fit these criteria were ultimately included in the study group. This included 27 patients of endometrial carcinoma, and 28 with benign masses (including 9 each of submucosal leiomyoma and endometrial hyperplasia, 5 with endometrial polyp and 5 with focal adenomyoma). The patients underwent multiparametric MRI and subsequent histopathological confirmation in the form of either endometrial biopsy or hysterectomy.

MRI protocol

All MRI examinations were performed using a 1.5 T scanner (Philips Achieva, Netherlands) using 32-channel surface coil with patient in supine position. Patients were asked to fast for at least 4 h prior to the examination and to void before the examination to reduce full-bladder compression of the uterus. Intravenous injection of 20 mg Hyoscine butyl bromide was given prior to the study to reduce bowel related motion artifacts.

The MRI protocol consisted of the following sequences:

1. T₂ weighted turbo spin echo DIXON in sagittal, axial oblique and coronal oblique planes.

2. T₁ weighted DIXON in axial oblique plane.

3. DWI in sagittal as well as axial oblique plane using b-values of 0, 500, 800 and 1000 s mm⁻²; ADC maps were derived automatically on a voxel by voxel basis using in-line software provided by the vendor during image generation.

4. Dynamic contrast-enhanced fat-saturated T₁ weighted three-dimensional gradient in the sagittal plane after injection of Gadoterate meglumine. These were taken as a part of a larger study.

5. Post-contrast T₁ weighted spin echo fat-saturated three-dimensional sequence.

The axial and coronal oblique sequences were performed along the long uterine axis. All sequences were performed in free breathing state. The complete protocol took 35–40 min in all cases.

MR image analysis

Two radiologists, one a genitourinary subspecialist with more than 10 years of experience and a fourth-year resident, independently measured the ADC values of the mass and the reference sites for internal control. Three readings of ADC in the endometrial mass, myometrium and gluteus maximus were obtained. The ratio of ADC of mass relative to normal myometrium and gluteus maximus muscle are hereby referred to as nADC_m and nADC_g respectively.

The method used for ADC measurement was consistent with that of previous authors.^7,10 For the mass, largest freehand region of interest (ROI) that could be drawn within the mass was selected. Uninvolved endometrium and necrotic areas were excluded from the ROIs on the basis of T₂ and post-contrast images (Figure 1). As reproducible localization is important for drawing ROI in the internal controls, the location of ROI in control organs was predefined. For gluteus maximus, it was kept in the center area of the gluteus maximus at the level of the inferior border of the sacroiliac joint. An exact site for the myometrium couldn’t be defined in view of variable infiltration of the myometrium by the mass. Hence, ROI was selected within normal appearing myometrium after careful correlation with other sequences, preferably in the same section as the mass. The size of the ROI in the internal controls was variable; the largest possible freehand ROI was drawn in the area of myometrium and gluteus maximus, excluding heterogeneous areas such as fatty infiltration of muscle or tumor infiltration of myometrium.

Figure 1.

Method of ROI placement for ADC measurement in a case of endometrial carcinoma. The layout on the console showing synced sagittal dynamic eTHRIVE 120 s post-contrast (a), T₂ weighted (b), DWI trace image at b-value 1000 (c) and ADC map (d). The section which best showed the tumour, avoiding necrosis and inhomogeneous areas within the tumour, with clear depiction of myometrium was chosen for measuring the ADC value of mass and myometrium by correlating with other sequences(d) Image showing ROI drawn in the mass and myometrium; the largest freehand ROI that could be drawn within the mass lesion was selected, taking care not to involve artifact from the tumour–air interface or blood flow. Another ROI selected in the adjacent normal appearing myometrium in the same section. Three measurements for mean ADC were taken in a similar fashion from different sections. The mean ADC of mass in this case was noted to be 0.66 × 10⁻³ mm² s⁻¹ and 1.64 × 10⁻³ mm² s⁻¹ for the myometrium. Axial ADC map at the level of gluteus maximus (e) demonstrates the measurement of ADC using largest freehand ROI. The level of gluteus maximus was fixed for all cases at the level of inferior SI joint as seen in coronal T₂ weighted (f) image level of inferior SI joint after ensuring proper muscle bulk. ADC, Apparent diffusion coefficient; DWI, Diffusion-weighted imaging; ROI, Region of interest

Histopathological analysis

All patients underwent histological examination (either endometrial biopsy or hysterectomy) 1–20 days after MRI examination. All females in the endometrial carcinoma group underwent hysterectomy (n = 27) while most patients in the benign group underwent either hysteroscopic resection (n = 10) or endometrial biopsy (n = 18). The surgical specimens were evaluated by a pathologist with 15 years of experience in gynecologic pathology who was blinded to the imaging results. In the endometrial cancer group, the histologic type (endometrioid or non-endometrioid) and tumor grade (G1, well-differentiated; G2, moderately differentiated; G3, poorly differentiated) was noted.

Statistical analysis

The intraclass correlation coefficient (ICC) was used to assess the interobserver agreement for ADC of endometrial mass, myometrium and gluteus maximus. The reproducibility between observers was considered to be good when the ICC was greater than 0.800. ANOVA was used to compare the ADC and nADC values in the different endometrial masses. Receiver operating characteristics (ROC) analysis was performed and the optimal cut-off value was extracted for ADC, nADC_m and nADC_g in differentiating benign and malignant endometrial masses. All analysis was performed with IBM SPSS Statistics (v. 22.0. Armonk, NY).

Results

A total of 55 females were recruited for the study. The age range was from 27 to 75 years with mean age of 53 years; mean age in the benign group was 46 years, compared to the malignancy group wherein the mean age was 59 years.

Interobserver agreement in ADC measurements

The interobserver agreement regarding the ADC of the endometrial masses and the control sites are summarized in Table 1. There was good interobserver agreement (>0.800) for all three sites; it was the highest for the endometrial lesion [ICC-0.98; CI (0.97–0.99)], while amongst the controls, it was higher for myometrium [ICC-0.92; CI (0.86–0.95)] as compared to gluteus maximus muscle [ICC-0.84; CI (0.72–0.90)] as elucidated in Table 1.

Table 1.

Interobserver agreement in ADC measurements

Organ	Mean ADC Reader 1	Mean ADC Reader 2 (×10 −3 mm2 s−1)	ICC	95% CI of ICC
Endometrial lesion	1.066 ± 0.363	1.065 ± 0.345	0.981	0.968–0.989
Myometrium	1.469 ± 0.156	1.515 ± 0.165	0.916	0.857–0.951
Gluteus maximus	1.167 ± 0.101	1.163 ± 0.184	0.838	0.723–0.905

ADC, Apparent diffusion coefficient; CI, Confidence interval;ICC, Intraclass correlation coefficient.

Mean ADC, nADC_m and nADC_g in benign and malignant endometrial masses

The mean ADC, nADC_m and nADC_g for different endometrial and subendometrial pathologies is depicted in Table 2. There was a statistically significant difference (p-0.02) in the mean ADC of endometrial carcinoma (0.77 ± 0.12 × 10⁻³ mm² s⁻¹) and benign masses (1.35 ± 0.24 x 10⁻³ mm² s⁻¹). The nADC_m in the malignant (0.52 ± 0.09) and benign (0.91 ± 0.19) masses showed statistically significant difference (p-0.02). The nADC_g in the malignant (0.69 ± 0.13) and benign (1.14 ± 0.24) masses also demonstrated statistically significant difference (p < 0.0001) as enumerated in Table 3 and Figure 2.

Table 2.

Mean ADC, nADC_m and nADC_g in different endometrial pathologies

Group	Mean ADC (×10−3 mm2 s−1)	Normalized ADCm	Normalized ADCg
Endometrial carcinoma (n = 27)	0.77 ± 0.12	0.52 ± 0.09	0.69 ± 0.13
Submucosal leiomyoma (n = 9)	1.34 ± 0.32	0.91 ± 0.17	1.10 ± 0.31
Hyperplasia (n = 9)	1.46 ± 0.11	0.93 ± 0.15	1.26 ± 0.16
Adenomyoma (n = 5)	1.09 ± 0.16	0.77 ± 0.19	0.97 ± 0.23
Polyp (n = 5)	1.06 ± 0.35	1.01 ± 0.21	1.16 ± 0.10

ADC, Apparent diffusion coefficient; nADC_g, Ratio of ADC of mass to gluteus maximus; nADC_m, Ratio of ADC of mass to myometrium.

Table 3.

Comparison of mean ADC, nADC_m and nADC_g in benign and malignant masses

Group	Mean ADC (×10−3 mm2 s−1) (p-0.02)	Normalized ADCm (p-0.02)	Normalized ADCg (p < 0.0001)
Malignant (n = 27)	0.77 ± 0.12	0.52 ± 0.09	0.69 ± 0.13
Benign (n = 28)	1.35 ± 0.24	0.91 ± 0.19	1.14 ± 0.24

ADC, Apparent diffusion coefficient; nADC_g, Ratio of ADC of mass to gluteus maximus; nADC_m, Ratio of ADC of mass to myometrium.

Figure 2.

Box and whisker plots of Mean ADC, nADCm and nADCg comparing benign and malignant endometrial masses. ADC, Apparent diffusion coefficient; nADCg, Ratio of ADC of mass to gluteus maximus; nADCm, Ratio of ADC of mass to myometrium.

ROC curve analysis (Figure 3) to establish cut-off to distinguish between benign and malignant masses showed that the cut-off value for mean ADC of 0.972 × 10⁻³ mm² s⁻¹ showed a sensitivity of 92.6% and a specificity of 92.9%, nADC_m of 0.652 showed a sensitivity of 92.6% and specificity of 92.9% and nADC_g of 0.815 showed a sensitivity of 88.9% and specificity of 92.9%.

Figure 3.

Receiver operating characteristics curve for the ADC, nADCm and nADCg for the differentiation of benign and malignant masses. ADC, Apparent diffusion coefficient; nADCg, Ratio of ADC of mass to gluteus max; nADCm, Ratio of ADC of mass to myometrium; ROC, Receiver operating characteristic

Tumor grade

There was no statistically significant difference in the mean ADC (p-0.068) and normalized ADC_g (p-0.166) of Grade 1, 2 and 3 of endometrioid type of carcinoma. However, normalized ADC_m (p-0.009) showed a statistically significant difference between the different tumor grades as shown in Table 4.

Table 4.

Mean ADC, nADC_m and nADC_g in different tumor grades and subtypes

Group	Mean ADC (×10−3 mm2 s−1)	Normalized ADCm	Normalized ADCg
Tumor grade	(p-0.068)	(p-0.009)	(p-0.166)
Endometrioid Grade 1 (n = 11)	0.79 ± 0.16	0.53 ± 0.12	0.73 ± 0.16
Endometrioid Grade 2 (n = 5)	0.71 ± 0.05	0.49 ± 0.08	0.65 ± 0.10
Endometrioid Grade 3 (n = 4)	0.70 ± 0.03	0.49 ± 0.04	0.59 ± 0.04
Tumor subtype	(p-0.50)	(p-0.625)	(p-0.813)
Type 1 (Endometrioid) (n = 20)	0.76 ± 0.13	0.51 ± 0.10	0.68 ± 0.14
Type 2 (Non-endometrioid) (n = 7)	0.79 ± 0.10	0.54 ± 0.07	0.68 ± 0.13

ADC, Apparent diffusion coefficient; nADC_g, Ratio of ADC of mass to gluteus maximus; nADC_m, Ratio of ADC of mass to myometrium.

Tumor subtype

Majority (20) of the patients (74%) had Type I (Endometrioid) carcinoma, while 7 patients (26%) had Type II (Non-endometrioid) carcinoma including five patients with serous type and 2 with carcinosarcoma. The difference in the mean ADC (p-0.50), normalized ADC_m(p-0.625), normalized ADC_g (p-0.813) were not statistically significant in the two groups as shown in Table 4.

Discussion

In this study, we evaluated the performance of a novel ADC parameter obtained by normalizing the ADC of endometrial lesion with myometrial ADC (nADC_m) and gluteal muscle ADC (nADC_g), which showed excellent sensitivity and specificity for distinguishing benign and malignant endometrial pathologies. The concept of using normalized ADC using an internal reference that is expected to be more comparable across different patients is relatively recent. Few studies have assessed the role of normalized ADC in abdominal imaging and have shown improved reproducibility and less variability in ADC measurement. Past studies^13,14 showed that normalized ADC of liver had additive value in distinguishing benign from malignant liver lesions and to distinguish liver fibrosis from normal liver respectively. Park et al¹⁵ studied the role of DWI for diagnosing metastatic lymph nodes in cervical cancer and showed improved performance of normalized ADC over absolute ADC. Barral et al¹⁶ showed that normalizing ADC of focal pancreatic lesions using apparently normal adjacent pancreatic parenchyma as control provides better characterization than absolute ADC. To our knowledge, there are no previous studies on normalized ADC for assessing uterine masses.

We assessed the value of myometrium and gluteal muscle for an internal control. The advantage of using myometrium as an internal reference includes its proximity to endometrial mass which allows more convenient ROI placement within the same plane as the mass, reducing inter- and intraobserver variability in measurement of ADC. Also, it can be assumed that the myometrium being in the same imaging plane as our target endometrial lesion is subject to similar field heterogeneity effects. The microenvironment of myometrium is expected to remain constant with patient age, built and menstrual phase. However, there may be complete myometrial infiltration in patients with advanced endometrial malignancy, in which cases no site for internal control may be available. This warranted the search for an alternative internal control that may serve as a viable choice in such cases where myometrial reading is not available. Gluteal muscle serves as an inert control in patients with endometrial pathologies, remaining constant over age and menstrual phase. Fatty infiltration due to muscular atrophy may be a confounding factor, however this can be easily correlated using conventional T₁- and T₂ weighted images and ROI placement can be adjusted accordingly.

The difference in absolute mean ADC, nADC_m and nADC_g between benign and malignant groups was found to be statistically significant in the present study. This finding is consistent with other previously published studies in the literature which studied absolute mean ADC values.^6–10 This difference in ADC values can be explained secondary to a pleomorphism and larger cell diameter and higher cellularity in malignancies than in benign tumors.^6,7 However, it was observed that rarely leiomyomas may have low ADC values that overlap with malignancies, which may be explained by the variable degenerations associated with leiomyomas. In such cases, we found that correlation with conventional and post-contrast sequences helped in distinction of the two pathologies. Ill-defined, heterogeneous and T₂ hyperintense masses, which are hypoenhancing in the post-contrast images favor a diagnosis of malignancy while the typical feature of submucosal fibroids is the well-defined, whorled heterogeneous signal intensity on all sequences, with early progressive contrast enhancement, that parallels the myometrium.

ROC curve analysis to establish cut-off value to distinguish between benign and malignant masses revealed a value for mean ADC of 0.972 × 10⁻³ mm² s⁻¹ with a sensitivity of 92.6% and a specificity of 92.9%, nADC_m of 0.652 showed a sensitivity of 92.6% and specificity of 92.9% and nADC_g of 0.815 showed a sensitivity of 88.9% and specificity of 92.9%, showing that nADC values have comparable accuracy as mean ADC. Previous studies have suggested mean ADC cut-off values to predict malignancy in endometrial lesions which range from 1.05 to 1.28 × 10⁻³ mm² s⁻¹.^5,7,9,10 It is difficult to establish a universal threshold ADC value owing to variability in the technique of acquisition, post-processing and measurement of ADC as detailed below, which highlights the value of normalized ADC.

This study revealed that the mean absolute and normalized ADC values had an inverse relationship with tumor grade and subtype. We found statistically significant (p-0.009) difference in the nADC_m values of Grade 1, 2 and 3 of endometrioid subtype. Endometrial carcinomas may also be divided based on the histology into endometrioid (Type I) and non-endometrioid (Type II) tumors.¹⁷ In our study, we did not find a statistically significant difference in the absolute or normalized ADC values between the two groups. Bakir et al¹¹ studied absolute ADC values and found a significant difference in ADC values between Grade I and Grade III (p-0.006), and non-endometrioid tumors (p-0.003). They also found a significant difference between endometrioid and non-endometrioid tumors (p-0.022). Yan et al¹² found no significant difference in the mean ADC between the different tumor grades of endometrial cancer.

There are various factors which influence the measurement of ADC values including acquisition techniques such as magnet strength, b-values, breath-hold vs free-breathing technique, post-processing software and measurement techniques including position and size of ROI placement for measuring ADC values.^18,19 Inoue et al²⁰ addressed the variations in shape of ROI and concluded that it does not affect the interobserver variability in endometrial carcinoma. Ghosh et al²¹ showed that there was a statistically significant difference in the mean ADC values obtained from different post-processing softwares when the same baseline trace diffusion images were used for the ADC map generation. Multiple studies in literature have reported inconsistent results on the relationship between tumor grade and histological subtype and ADC. Use of different groups of tumor grades in the studies (such as Grade I + II vs Grade III, Grade I vs Grade II + III) to classify low grade from high grade malignancy is one of the factors leading to the variable results.

Our study has few limitations. Our results were obtained from a single center study with a limited sample size, reflecting our preliminary experience and lower-case burden for endometrial carcinoma in our country.²² There is a disparity in the number of patients in the endometrial carcinoma subgroup with higher number of patients (74%) with Type 1 endometrioid carcinoma. Larger multicenter studies comparing normalized ADC readings in different acquisition protocols and post-processing techniques are needed to prove its benefit. This study may serve as a preliminary investigation which reveals that normalized ADC is as worthwhile, if not more, as absolute ADC for characterization of benign and malignant endometrial masses.

Conclusion

This study shows that normalized ADC is useful to distinguish benign and malignant masses with comparable accuracy as absolute ADC. Normalized ADC may provide a useful, robust and more reproducible quantitative parameter which limits the influence of endogenous and exogenous factors that affect its accuracy and reproducibility.