Utility of T2-weighted MRI to Differentiate Adrenal Metastases from Lipid-Poor Adrenal Adenomas

Abstract

Purpose

To evaluate T2-weighted MRI features to differentiate adrenal metastases from lipid-poor adenomas.

Materials and Methods

With institutional review board approval, this study retrospectively compared 40 consecutive patients (mean age, 66 years ± 10 [standard deviation]) with metastases to 23 patients (mean age, 60 years ± 15) with lipid-poor adenomas at 1.5- and 3-T MRI between June 2016 and March 2019. A blinded radiologist measured T2-weighted signal intensity (SI) ratio (SI_nodule/SI_{psoas muscle}), T2-weighted histogram features, and chemical shift SI index. Two blinded radiologists (radiologist 1 and radiologist 2) assessed T2-weighted SI and T2-weighted heterogeneity using five-point Likert scales.

Results

Subjectively, T2-weighted SI (P < .001 for radiologist 1 and radiologist 2) and T2-weighted heterogeneity (P < .001, for radiologist 1 and radiologist 2) were higher in metastases compared with adenomas when assessed by both radiologists. Agreement between the radiologists was substantial for T2-weighted SI (Cohen κ = 0.67) and T2-weighted heterogeneity (κ = 0.62). Metastases had higher T2-weighted SI ratio than adenomas (3.6 ± 1.7 [95% confidence interval {CI}: 0.2, 8.2] vs 2.2 ± 1.0 [95% CI: 0.6, 4.3], P < .001) and higher T2-weighted entropy (6.6 ± 0.6 [95% CI: 4.9, 7.5] vs 5.0 ± 0.8 [95% CI: 3.5, 6.6], P < .001). At multivariate analysis, T2-weighted entropy was the best differentiating feature (P < .001). Chemical shift SI index did not differ between metastases and adenomas (P = .748). Area under the receiver operating characteristic curve (AUC) for T2-weighted SI ratio and T2-weighted entropy were 0.76 (95% CI: 0.64, 0.88) and 0.94 (95% CI: 0.88, 0.99). The logistic regression model combining T2-weighted SI ratio with T2-weighted entropy yielded AUC of 0.95 (95% CI: 0.91, 0.99) and did not differ compared with T2-weighted entropy alone (P = .268). There was no difference in logistic regression model accuracy comparing the data by either field strength, 1.5- or 3-T MRI (P > .05).

Conclusion

Logistic regression models combining T2-weighted SI and T2-weighted heterogeneity can differentiate metastases from lipid-poor adenomas. Validation of these preliminary results is required.

Keywords: Adrenal, MR-Imaging, Urinary

Supplemental material is available for this article.

© RSNA, 2020

Summary

Logistic regression models incorporating quantitative T2-weighted signal intensity ratio and T2-weighted heterogeneity are highly accurate to differentiate adrenal metastases from lipid-poor adrenal adenomas.

Key Points

• ■ Adrenal metastases have higher T2-weighted signal intensity (SI) compared with lipid-poor adrenal adenomas (P < .001).

• ■ Adrenal metastases have higher T2-weighted heterogeneity scores, assessed subjectively (P < .001) and quantitatively with higher T2-weighted entropy (P < .001), compared with lipid-poor adrenal adenomas.

• ■ Logistic regression models using T2-weighted SI with T2-weighted entropy (area under the receiver operating characteristic curve [AUC] of 0.95) or T2-weighted SI with subjective T2-weighted heterogeneity scores (AUC of 0.88 for radiologist 1 and AUC of 0.83 for radiologist 2) showed high accuracy for diagnosis of adrenal metastases.

Introduction

Adrenal nodules are common incidental imaging findings discovered in patients undergoing CT and MRI examinations for other clinical reasons. In patients with no history of malignancy, incidentally depicted small (< 4 cm) homogeneous adrenal nodules are almost always benign adrenal adenomas (1–6). Although the likelihood of malignancy in an incidental adrenal nodule is exceedingly rare, when there is a personal history of malignancy, the risk increases substantially (7). For this reason, incidentally discovered adrenal nodules in patients with a history of cancer require further characterization to establish a diagnosis.

In patients imaged with MRI, the diagnosis of adrenal adenoma is predicated on the ability to document microscopic fat in lipid-rich adenomas by either subjective analysis or by using quantitative signal intensity (SI) ratios measured at dual-echo chemical shift MRI (4,8–13). Chemical shift MRI is the most sensitive examination for documentation of microscopic fat in adrenal adenomas and will characterize an increased proportion of adrenal nodules considered lipid poor measuring between 10 and 30 HU at unenhanced CT (14–16). Nevertheless, it is estimated that approximately 30% of adrenal adenomas are lipid poor (ie, contain insufficient amounts of fat within the cytoplasm of adrenal cortical cells to be detected with imaging), and these will be indeterminate even at chemical shift MRI (15,17). Studies comparing MRI and adrenal washout CT for characterization of adrenal nodules measuring 10 HU in attenuation have shown superiority of washout CT; however, they are limited by a heterogeneous control group of “nonadenomas” which include as few as four metastases (18,19). Moreover, adrenal washout CT may be limited to differentiate adenomas from hypervascular metastases (8).

Previous investigators have shown that adenomas tend to be of lower T2-weighted SI compared with metastases, pheochromocytoma, and adrenal cortical carcinoma (10,20–22) and also more homogeneous (23). Investigators have also more recently shown the value of quantitative texture analysis in adrenal nodules imaged with CT and chemical shift MRI (24); however, to our knowledge they have not quantitatively studied SI and texture analysis of adrenal nodules at T2-weighted MRI to differentiate lipid-poor adrenal adenomas from metastases. Our hypothesis is that lipid-poor adrenal adenomas have lower T2-weighted SI and are more homogeneous at T2-weighted MRI than metastases, which may enable accurate characterization. The purpose of this study was therefore to evaluate the utility of T2-weighted MRI SI and T2-weighted heterogeneity measurements to differentiate between lipid-poor adrenal adenomas and adrenal metastases.

Materials and Methods

Study Design

We determined a priori that a sample of at least 20 metastases and 20 lipid-poor adenomas was required to enable modeling that would include the two features evaluated in the present study (T2-weighted SI ratio and T2-weighted heterogeneity). It has been previously shown using the “rule of 10” that a minimum of 10 events are needed per feature included in modeling to reduce the risk of overfitting of a model (25). These features were selected a priori based on our own clinical observation and prior data showing that adrenal adenomas have lower T2-weighted SI and are more homogeneous compared with other adrenal masses (10,20–22,26).

With institutional review board approval that waived the need for informed consent in all patients, we performed a retrospective search of reports using a picture archiving and communication system search tool (McKesson Study Share; McKesson, Irving, Tex) from a single institution between the dates of June 26, 2016, and March 1, 2019, for the terms “adrenal metastasis” or “adrenal metastases” under the search filter “MRI” as inclusion criteria. We identified 52 patients who met the search criteria. Twelve patients were excluded due to the size of the adrenal nodule (< 1 cm in size) (Fig 1). A size threshold of 1 cm was selected in accordance with the American College of Radiology Incidental Findings Committee and due to the limitations in spatial resolution at MRI precluding accurate quantitative assessment of nodules of smaller than 1 cm on chemical shift and T2-weighted images (27,28). Each patient in the search had a single adrenal metastasis which was indeterminate at time of first imaging. All patients already had a diagnosis of cancer at time of MRI; metastases from primary malignancies included in this study were from lung (n = 16), colon (n = 4), hepatocellular carcinoma (n = 6), renal cell carcinoma (n = 7), ovarian (n = 1), breast (n = 3), pancreatic (n = 2), and gastric cancer (n = 1). A diagnosis of adrenal metastasis was established using the following three criteria: (a) histologic confirmation from adrenalectomy specimen or 18-gauge core needle biopsy (n = 7), (b) interval development of an adrenal nodule when compared with a preceding CT or MRI examination which showed a previously normal adrenal gland with no nodule and subsequent growth on follow-up imaging studies (n = 19), and (c) short-term interval (< 6 month) growth of the nodule in the context of progression of metastatic disease (new metastases or > 20% increase in the sum of disease [29]) elsewhere in the same patient (n = 14). The mean percentage increase in size of adrenal nodules (measured by dividing the difference in measurements between scans divided by the initial measurement) on follow-up was 40.7% ± 19.1 (range, 13.2%–73.3%) with mean time interval between scans of 398 days ± 956 (range, 12–3802 days). Patient demographics are summarized in Table 1.

Figure 1:

Flow diagram of patient inclusion and exclusion in the present study. Lipid-rich and lipid-poor adenomas were differentiated by chemical shift signal intensity index and unenhanced CT attenuation threshold of 16.5% and 10 HU, respectively.

Table 1:

Patient Demographics for Adrenal Metastases and Adrenal Adenomas

The control group was also created by performing a similar retrospective search of reports using the same search tool between the same study dates for terms “lipid-poor adrenal nodule OR adrenal adenoma OR adrenal adenomas” also under the search filter “MRI.” A total of 76 patients were identified with potential lipid-poor adrenal adenomas (Fig 1). To be considered a lipid-poor adenoma, chemical shift SI index measurements should be less than 16.5% and unenhanced CT measurements more than 10 HU (13). For diagnosis of lipid-poor adenomas, follow-up imaging performed after at least a 6-month interval showed no change in size of the nodule with no intervening treatment and with negative biochemical testing. Fifty-three adrenal adenomas were excluded for being lipid rich (chemical shift SI index > 16.5% and unenhanced CT < 10 HU) or because of insufficient reference standards. A total of 23 patients were identified for the control group. Mean time to follow-up for establishing stability in size in this study was 1902 days ± 1066 (range, 372–4842 days). Patient demographics are also provided in Table 1. A total of 26.1% (six of 23) of patients from the adrenal adenoma cohort had been previously studied by investigators to compare MRI features to pheochromocytoma (20); however, comparison of lipid-poor adenomas to metastases has not been previously studied by our group.

MRI Technique

In 93.7% (59 of 63) of patients, MRI was performed at a single center using one of three clinical 1.5-T (58.7% [37 of 63]) or 3-T (34.9% [22 of 63]) systems (Symphony or TRIO, Siemens Healthcare, Malvern, Pa; or Discovery 750 W, GE Healthcare, Milwaukee, Wis). Pulse sequence parameters are provided in Table 2. The other 6.3% (four of 63) of patients underwent MRI using 1.5-T clinical scanners (Achieva, Philips, Andover, Mass; and Avanto or Aera, Siemens Healthcare) with similar imaging sequences described above, performed at outside institutions, with images available to be reviewed using our institutional picture archiving and communication system.

Table 2:

Multiparametric MRI Technique Used for Adrenal Masses at Our Institution

MRI Region of Interest Analysis

A radiologist, blinded to the diagnosis but with knowledge of where the adrenal nodule was located in each patient, with 4 years of experience in abdominal MRI (W.T.), independently measured SI values. Measurements were performed on axial T2-weighted images, selecting the center slice of the nodule where it appeared the largest. For homogeneous nodules, a circular region of interest (ROI) was placed within the nodule encompassing at least two-thirds of the lesion (Fig 2). For heterogeneous nodules, a circular ROI was placed within the nodule encompassing the most T2-weighted hyperintense aspect of the nodule, judged subjectively, and measuring at least 5 mm in diameter (Fig 3). This measurement technique, described previously, accounts for potential averaging of SI values in heterogeneous nodules (30–32). A fixed diameter (5 mm) circular ROI was placed in the ipsilateral psoas muscle so that the T2-weighted nodule-to-muscle SI ratio could be calculated: (SI_nodule/SI_muscle) (33). The psoas muscle was used as the internal reference standard at T2-weighted MRI, as measurements could be made at roughly the same anteroposterior level as the adrenal mass to minimize changes related to receiver coil design (34). At time of SI measurement on T2-weighted images, the radiologist also measured the size of each nodule in three planes with the mean size recorded in millimeters.

Figure 2a:

MR images in a 48-year-old woman with right lipid-poor adrenal adenoma. (a) Axial T2-weighted single-shot turbo spin-echo image depicts the right adrenal nodule (arrow) with homogeneously low signal intensity (SI). (b) Inset image of (a) shows method of measurement of T2-weighted SI ratio and histogram features. Circular (yellow) region of interest (ROI) depicts method of measurement of adrenal T2-weighted SI for homogeneous nodules. A similar-sized ROI was placed on the ipsilateral psoas muscle (not shown) to measure the adrenal-to-muscle T2-weighted SI ratio. Custom fit ROI (white) shows method of contouring of adrenal masses for extraction of histogram features. (c, d) Axial T1-weighted in-phase (IP) and opposed-phase (OP) gradient recalled-echo images show that there is no SI drop on OP compared with IP images (arrow), indicating the absence of microscopic fat. Chemical shift SI index was less than 16.5%. The diagnosis of lipid-poor adrenal adenoma was established by CT washout and negative biochemical testing. The nodule was stable for 2 years in follow-up.

Figure 3a:

MR images in a 67-year-old woman with left adrenal metastasis from lung cancer. (a) Axial T2-weighted half-Fourier acquired single-shot turbo spin-echo (HASTE) image shows the left adrenal mass as having heterogeneously increased signal intensity (SI) (arrow). (b) Axial T2-weighted HASTE image depicting method of region of interest (ROI) placement for heterogeneous nodules. Circular ROI (black) depicts method of measurement of T2-weighted SI for heterogeneous nodules; the most T2-weighted hyperintense area was measured using an ROI that measured at least 5 mm in diameter. Custom fit ROI (white circle) shows method of contouring of adrenal masses for extraction of histogram features. (c, d) Axial T1-weighted in-phase (IP) and opposed-phase (OP) gradient recalled-echo images show no loss of SI comparing OP to IP (arrow), indicating the absence of microscopic fat. Chemical shift SI index was less than 16.5%. Diagnosis was established by a doubling in size at 3-month follow-up CT.

Figure 2b:

MR images in a 48-year-old woman with right lipid-poor adrenal adenoma. (a) Axial T2-weighted single-shot turbo spin-echo image depicts the right adrenal nodule (arrow) with homogeneously low signal intensity (SI). (b) Inset image of (a) shows method of measurement of T2-weighted SI ratio and histogram features. Circular (yellow) region of interest (ROI) depicts method of measurement of adrenal T2-weighted SI for homogeneous nodules. A similar-sized ROI was placed on the ipsilateral psoas muscle (not shown) to measure the adrenal-to-muscle T2-weighted SI ratio. Custom fit ROI (white) shows method of contouring of adrenal masses for extraction of histogram features. (c, d) Axial T1-weighted in-phase (IP) and opposed-phase (OP) gradient recalled-echo images show that there is no SI drop on OP compared with IP images (arrow), indicating the absence of microscopic fat. Chemical shift SI index was less than 16.5%. The diagnosis of lipid-poor adrenal adenoma was established by CT washout and negative biochemical testing. The nodule was stable for 2 years in follow-up.

Figure 2c:

MR images in a 48-year-old woman with right lipid-poor adrenal adenoma. (a) Axial T2-weighted single-shot turbo spin-echo image depicts the right adrenal nodule (arrow) with homogeneously low signal intensity (SI). (b) Inset image of (a) shows method of measurement of T2-weighted SI ratio and histogram features. Circular (yellow) region of interest (ROI) depicts method of measurement of adrenal T2-weighted SI for homogeneous nodules. A similar-sized ROI was placed on the ipsilateral psoas muscle (not shown) to measure the adrenal-to-muscle T2-weighted SI ratio. Custom fit ROI (white) shows method of contouring of adrenal masses for extraction of histogram features. (c, d) Axial T1-weighted in-phase (IP) and opposed-phase (OP) gradient recalled-echo images show that there is no SI drop on OP compared with IP images (arrow), indicating the absence of microscopic fat. Chemical shift SI index was less than 16.5%. The diagnosis of lipid-poor adrenal adenoma was established by CT washout and negative biochemical testing. The nodule was stable for 2 years in follow-up.

Figure 2d:

MR images in a 48-year-old woman with right lipid-poor adrenal adenoma. (a) Axial T2-weighted single-shot turbo spin-echo image depicts the right adrenal nodule (arrow) with homogeneously low signal intensity (SI). (b) Inset image of (a) shows method of measurement of T2-weighted SI ratio and histogram features. Circular (yellow) region of interest (ROI) depicts method of measurement of adrenal T2-weighted SI for homogeneous nodules. A similar-sized ROI was placed on the ipsilateral psoas muscle (not shown) to measure the adrenal-to-muscle T2-weighted SI ratio. Custom fit ROI (white) shows method of contouring of adrenal masses for extraction of histogram features. (c, d) Axial T1-weighted in-phase (IP) and opposed-phase (OP) gradient recalled-echo images show that there is no SI drop on OP compared with IP images (arrow), indicating the absence of microscopic fat. Chemical shift SI index was less than 16.5%. The diagnosis of lipid-poor adrenal adenoma was established by CT washout and negative biochemical testing. The nodule was stable for 2 years in follow-up.

Figure 3b:

MR images in a 67-year-old woman with left adrenal metastasis from lung cancer. (a) Axial T2-weighted half-Fourier acquired single-shot turbo spin-echo (HASTE) image shows the left adrenal mass as having heterogeneously increased signal intensity (SI) (arrow). (b) Axial T2-weighted HASTE image depicting method of region of interest (ROI) placement for heterogeneous nodules. Circular ROI (black) depicts method of measurement of T2-weighted SI for heterogeneous nodules; the most T2-weighted hyperintense area was measured using an ROI that measured at least 5 mm in diameter. Custom fit ROI (white circle) shows method of contouring of adrenal masses for extraction of histogram features. (c, d) Axial T1-weighted in-phase (IP) and opposed-phase (OP) gradient recalled-echo images show no loss of SI comparing OP to IP (arrow), indicating the absence of microscopic fat. Chemical shift SI index was less than 16.5%. Diagnosis was established by a doubling in size at 3-month follow-up CT.

Figure 3c:

MR images in a 67-year-old woman with left adrenal metastasis from lung cancer. (a) Axial T2-weighted half-Fourier acquired single-shot turbo spin-echo (HASTE) image shows the left adrenal mass as having heterogeneously increased signal intensity (SI) (arrow). (b) Axial T2-weighted HASTE image depicting method of region of interest (ROI) placement for heterogeneous nodules. Circular ROI (black) depicts method of measurement of T2-weighted SI for heterogeneous nodules; the most T2-weighted hyperintense area was measured using an ROI that measured at least 5 mm in diameter. Custom fit ROI (white circle) shows method of contouring of adrenal masses for extraction of histogram features. (c, d) Axial T1-weighted in-phase (IP) and opposed-phase (OP) gradient recalled-echo images show no loss of SI comparing OP to IP (arrow), indicating the absence of microscopic fat. Chemical shift SI index was less than 16.5%. Diagnosis was established by a doubling in size at 3-month follow-up CT.

Figure 3d:

MR images in a 67-year-old woman with left adrenal metastasis from lung cancer. (a) Axial T2-weighted half-Fourier acquired single-shot turbo spin-echo (HASTE) image shows the left adrenal mass as having heterogeneously increased signal intensity (SI) (arrow). (b) Axial T2-weighted HASTE image depicting method of region of interest (ROI) placement for heterogeneous nodules. Circular ROI (black) depicts method of measurement of T2-weighted SI for heterogeneous nodules; the most T2-weighted hyperintense area was measured using an ROI that measured at least 5 mm in diameter. Custom fit ROI (white circle) shows method of contouring of adrenal masses for extraction of histogram features. (c, d) Axial T1-weighted in-phase (IP) and opposed-phase (OP) gradient recalled-echo images show no loss of SI comparing OP to IP (arrow), indicating the absence of microscopic fat. Chemical shift SI index was less than 16.5%. Diagnosis was established by a doubling in size at 3-month follow-up CT.

For chemical shift (in-phase [IP] and opposed-phase [OP]) MRI, a circular ROI was placed in the nodule at the same level on axial IP and OP images as was performed for T2-weighted images, avoiding the edges of the nodule to not include areas of India ink artifact (Figs 2, 3) (35). A fixed diameter (5 mm) ROI was also placed in the spleen. Chemical shift SI index

and chemical shift adrenal-to-spleen (ASR) SI ratio

were calculated as described previously (33,36,37).

MRI Histogram Analysis

For each tumor, the single axial image used for T2-weighted SI measurements was segmented manually using ImageJ software (version 1.52; National Institutes of Health, Bethesda, Md). Each axial image was manually contoured to define the outer margin of the nodule (Figs 2, 3). Two-dimensional histogram analysis was performed, and three parameters were studied. Kurtosis (a measure of histogram flatness) and skewness (a measure of histogram asymmetry) were extracted directly from ImageJ. Entropy (a measure of histogram irregularity, with larger values indicating increased heterogeneity of SI values) was obtained using: −sum (p.*log2 (p)), where p contains the histogram pixel SI values, through an in-house software plugin created for Microsoft Excel (version 14.0; Microsoft, Redmond, Wash). We studied two-dimensional first-order texture analysis (histogram) features, as these have been well studied and validated in adrenal and renal masses (32,38,39). Single-shot T2-weighted MRI was used to measure histogram features because it is a robust imaging sequence providing reliable image quality between patients with minimal artifact; it uses similar pulse sequence parameters, including voxel size, between systems and vendors; it is included in most abdominal MRI examinations; it does not rely on the injection of gadolinium-based contrast material; and it has been previously investigated and validated for the study of texture analysis features in renal and adrenal tumors with MRI (21,40–43). The process of exporting an anonymized Digital Imaging and Communications in Medicine (DICOM) image and performing segmentation and histogram analysis required approximately 5 minutes per patient and the use of a separate laptop operating ImageJ and Microsoft Excel.

Reproducibility of Quantitative Measurements

To calculate the reproducibility of our quantitative measurements, a second radiologist (also blinded to diagnosis but provided with the location of the adrenal nodule), with 7 years of experience in MRI (J.A.G.), segmented the same T2-weighted DICOM images for every patient.

MRI Subjective Analysis

Two radiologists, with 7 and 13 years of experience in MRI (J.A.G. and N.S., respectively), blinded to the final diagnosis but provided with the location of each adrenal nodule, independently evaluated the following features on T2-weighted images: (a) nodule homogeneity using a five-point Likert scale (Fig E2 [supplement]): (i) completely heterogeneous (areas of high and low SI distributed throughout the whole nodule), (ii) mostly heterogeneous, (iii) mixed areas consisting of both homogeneous and heterogeneous SI within the nodule, (iv) mostly homogeneous, and (v) completely homogeneous (completely uniform SI) and (b) nodule SI compared with ipsilateral renal cortex using a five-point Likert scale (Figs E1 and E2 [supplement]): (i) very hyperintense to renal cortex, (ii) slightly hyperintense to renal cortex, (iii) isointense to renal cortex, (iv) slightly hypointense to renal cortex, and (v) very hypointense to renal cortex. Subjective assessment of T2-weighted SI was performed relative to the renal cortex to provide a wider range for visual assessment, because in our experience most adrenal masses, including metastases and adenomas, are of increased signal relative to the psoas muscle.

Statistical Analysis

All parametric data are presented as mean ± standard deviation (with range also provided). Parametric data were tested for skewness (P > .05) to determine whether parametric or nonparametric testing was most appropriate. Demographic variables and subjective outcomes were compared using the χ² test of proportions and Wilcoxon signed rank tests. Parametric data were compared using Student t tests. Multivariate logistic regression modeling was then performed using T2-weighted SI, subjective T2-weighted heterogeneity scores, and the best performing quantitative T2-weighted histogram metric. The sample size enabled up to two features per model, minimizing the risk of overfitting (25); therefore, the models consisted of (a) T2-weighted signal intensity with T2-weighted entropy and (b) T2-weighted SI with T2-weighted subjective heterogeneity scores as determined a priori.

Receiver operating characteristic (ROC) analysis was performed to determine diagnostic accuracy. A subgroup analysis of the predictive models was compared by magnetic field strength and by adrenal nodules less than 4 cm. We explored the relationship between nodule size and T2-weighted entropy using Pearson correlation. The optimal diagnostic accuracy was determined using the method described by Youden. Comparison of accuracies between tests was performed using ROC comparison analysis. Threshold P value less than .05 indicated a statistically significant difference. Interobserver agreement was assessed for subjective Likert scoring using Cohen κ statistic and using Dice similarity coefficient for T2-weighted quantitative data. Statistical analysis was performed with Stata data analysis and statistical software, version 13 (Stata, College Station, Tex).

Results

Patient Overview

Adrenal metastases were larger than adenomas (34 mm ± 20 [range, 13–95 mm] vs 15 mm ± 4 [10–22 mm], P < .001) with otherwise no difference between sex (P = .681), age (P = .062), or laterality of adrenal involvement (P = .242). Patient demographics are summarized in Table 1.

Subjective Assessment

Comparison of subjective features assessed by both radiologists in lipid-poor adenomas and metastases are summarized in Table 3. Metastases were subjectively assessed as having higher T2-weighted SI (P < .001 for both radiologists 1 and 2) and were more heterogeneous on T2-weighted images (P < .001 for both radiologists). A Likert score of 1 or 2 optimized specificity for diagnosis of metastases, as no lipid-poor adenomas were considered to be slightly hyperintense or very hyperintense (Likert scores 1 and 2) relative to the renal cortex for either radiologist. Comparatively, radiologist 1 and 2 rated 45.0% (18 of 40) and 47.5% (19 of 40), respectively, of metastases as slightly or very hyperintense. A Likert score of 3 or less optimized specificity for diagnosis of metastases, as only one lipid-poor adenoma for radiologist 2 was rated as mostly or completely heterogeneous (Likert scores 1 and 2) compared with 50.0% (20 of 40) and 47.5% (19 of 40) of metastases which were considered mostly or completely heterogeneous by radiologist 1 and 2, respectively. A summary of diagnostic accuracy of subjective assessment of T2-weighted SI and T2-weighted heterogeneity and a logistic regression model combining T2-weighted SI and heterogeneity for each radiologist is provided in Table 4. Agreement was substantial for both T2-weighted SI (κ = 0.67) and T2-weighted heterogeneity scoring (κ = 0.62) between radiologists.

Table 3:

Subjective Evaluation of T2-weighted Signal Intensity and Heterogeneity

Table 4:

Diagnostic Accuracy for Diagnosis of Adrenal Metastases from Lipid-Poor Adenomas

Quantitative Assessment

Comparisons of quantitative features in lipid-poor adenomas and metastases are summarized in Table 5. Reproducibility of segmentations between radiologists was good (Dice score coefficient = 0.69). There was no difference in measured T2-weighted SI ratio or T2-weighted histogram features comparing the measurements performed by the two radiologists (P > .05). Adrenal metastases had higher values than lipid-poor adenomas for T2-weighted SI ratio (3.6 ± 1.7 vs 2.2 ± 1.0; P < .001), T2-weighted entropy (6.6 ± 0.6 vs 5.0 ± 0.8; P < .001), T2-weighted skewness (0.5 ± 0.8 vs −0.01 ± 0.5; P = .002), and T2-weighted kurtosis (1.0 ± 2.8 vs −0.2 ± 0.6; P = .036). At T1-weighted dual-echo chemical shift MRI, there was no difference in chemical shift SI index or adrenal-to-spleen SI ratio comparing metastases and lipid-poor adenomas (P = .748 and .854). Of note, 7.5% (three of 40) of metastases had chemical shift SI index in the adenoma range (> 16.5%) from hepatocellular carcinoma, colon, and lung primary malignancies (Fig 4). By definition, no lipid-poor adenomas had chemical shift SI index of greater than 16.5%.

Table 5:

Summary of Quantitative Comparisons between Lipid-Poor Adrenal Adenomas and Adrenal Metastases

Figure 4a:

MR images in a 72-year-old man with right adrenal metastasis from hepatocellular carcinoma. (a) Axial T2-weighted half-Fourier acquired single-shot turbo spin-echo image shows a heterogeneous right adrenal nodule (arrow) with increased signal intensity (SI) relative to renal cortical parenchyma. Note simple cyst in the right kidney (*). (b, c) Axial T1-weighted in-phase (IP) and opposed-phase (OP) gradient recalled-echo images show heterogeneous loss of SI comparing OP to IP within the nodule (arrow) indicating the presence of microscopic fat. The nodule was confirmed to represent a metastasis as it was newly developed (not present on preceding CT scan performed 1 year prior) and increased by 32% in size at follow-up CT performed 6 months later (not shown).

Figure 4b:

MR images in a 72-year-old man with right adrenal metastasis from hepatocellular carcinoma. (a) Axial T2-weighted half-Fourier acquired single-shot turbo spin-echo image shows a heterogeneous right adrenal nodule (arrow) with increased signal intensity (SI) relative to renal cortical parenchyma. Note simple cyst in the right kidney (*). (b, c) Axial T1-weighted in-phase (IP) and opposed-phase (OP) gradient recalled-echo images show heterogeneous loss of SI comparing OP to IP within the nodule (arrow) indicating the presence of microscopic fat. The nodule was confirmed to represent a metastasis as it was newly developed (not present on preceding CT scan performed 1 year prior) and increased by 32% in size at follow-up CT performed 6 months later (not shown).

Figure 4c:

MR images in a 72-year-old man with right adrenal metastasis from hepatocellular carcinoma. (a) Axial T2-weighted half-Fourier acquired single-shot turbo spin-echo image shows a heterogeneous right adrenal nodule (arrow) with increased signal intensity (SI) relative to renal cortical parenchyma. Note simple cyst in the right kidney (*). (b, c) Axial T1-weighted in-phase (IP) and opposed-phase (OP) gradient recalled-echo images show heterogeneous loss of SI comparing OP to IP within the nodule (arrow) indicating the presence of microscopic fat. The nodule was confirmed to represent a metastasis as it was newly developed (not present on preceding CT scan performed 1 year prior) and increased by 32% in size at follow-up CT performed 6 months later (not shown).

Area under ROC curve (AUC) for diagnosis of metastasis using T2-weighted SI ratio and T2-weighted entropy evaluated independently were 0.76 (95% confidence interval [CI]: 0.64, 0.88) and 0.94 (95% CI: 0.88, 0.99) (Fig 5), respectively. Multivariable logistic regression showed differences for T2-weighted SI (P = .089), entropy (P < .001), and T2-weighted SI plus entropy (P < .001). AUC for logistic regression model combining T2-weighted SI ratio and T2-weighted entropy was 0.95 (95% CI: 0.91, 0.99) with an optimal sensitivity of 75.0% and specificity of 100.0% (Fig 4). AUCs for logistic regression model combining T2-weighted SI ratio with subjective heterogeneity scores were 0.88 (95% CI: 0.79, 0.86) for radiologist 1 and 0.83 (95% CI: 0.73, 0.93) for radiologist 2, with an optimal sensitivity and specificity of 62% and 100% for radiologist 1, respectively, and 55.0% and 100% for radiologist 2, respectively (Fig 4). The logistic regression model combining T2-weighted SI ratio and entropy outperformed the other models tested, including using subjective assessment alone or in combination with T2-weighted SI measurements, (P < .001–.05); however, they did not differ compared with accuracy of T2-weighted entropy alone (P = .268).

Figure 5:

Receiver operating characteristic curves comparing T2-weighted signal intensity (SI) ratio (dotted line), T2-weighted entropy (long dashed line), logistic regression model of T2-weighted SI ratio with T2-weighted entropy (solid line), logistic regression model of T2-weighted SI ratio with subjective heterogeneity scores by radiologist 1 (dash and dot line), and logistic regression model of T2-weighted SI ratio with subjective heterogeneity scores by radiologist 2 (tight dashed line). FPR = false positive rate, TPR = true positive rate.

We performed two subgroup analyses to internally validate our results. A subgroup analysis of the logistic regression model combining T2-weighted SI ratio and T2-weighted entropy at 1.5 T (n = 41) and 3 T (n = 22) yielded AUCs of 0.97 (95% CI: 0.91, 1.0) and 0.95 (95% CI: 0.87, 1.0), respectively, with no difference comparing AUCs at either field strength (P > .05). Moreover, a subgroup analysis evaluating masses less than 4 cm in size comparing metastases (n = 29, 24 mm ± 6) to lipid-poor adenomas (n = 23, 15 mm ± 4) showed similar high accuracy of classification of metastases using the logistic regression model combining T2-weighted SI ratio and T2-weighted entropy with AUC of 0.94 (95% CI: 0.88, 1.00). There was a moderate positive correlation between nodule size and T2-weighted entropy (κ = .62, P < .001).

Discussion

This study evaluated the ability of T2-weighted MRI to differentiate adrenal metastases from lipid-poor adrenal adenomas using SI and heterogeneity as potentially discriminative imaging features. In our study, adrenal metastases were more hyperintense and more heterogeneous on T2-weighted images compared with adrenal adenomas; heterogeneity, in particular, differed most between the two groups. Models combining T2-weighted SI ratio with tumor heterogeneity, assessed quantitatively (T2-weighted entropy) or subjectively (five-point Likert scoring), were highly accurate to differentiate metastases from lipid-poor adenomas. Our results indicate that in a patient with known malignancy and an indeterminate adrenal nodule not containing microscopic fat at MRI, heterogeneously increased T2-weighted SI favors a diagnosis of metastatic disease and may help to prevent further unnecessary confirmatory testing such as adrenal washout CT, PET, or biopsy. Validation of these results is required.

Varghese et al were the first to evaluate adrenal nodule SI at T2-weighted MRI comparing pheochromocytomas and adenomas, demonstrating that a majority of adenomas showed low T2-weighted signal (44). More recent studies evaluating the SI of adenomas compared with metastases both qualitatively (22) and quantitatively (21) confirmed that adenomas have lower T2-weighted SI compared with metastases. In our study, T2-weighted SI was also significantly lower in lipid-poor adenomas compared with a large group of adrenal metastases from various primary malignancies using both subjective and quantitative assessment with a five-point Likert scale and through the measurement of a simple T2-weighted SI ratio.

The use of texture analysis in adrenal assessment has been predominantly applied in CT, with few studies evaluating texture features of adrenal masses on MR images (24,45–50). In our study, both qualitative and quantitative analyses of tumor heterogeneity revealed that metastases were significantly more heterogeneous compared with lipid-poor adenomas, which is comparable to a prior study which compared adrenal metastases from renal cell carcinoma and adenomas at T2-weighted MRI (21). T2-weighted SI and heterogeneity may be very useful in cases of renal cell carcinoma and hepatocellular carcinomas, as metastases may show microscopic fat simulating lipid-rich adenomas at chemical shift MRI (8,51) and because both tumors are hypervascular and may wash out into the adenoma range at washout CT examinations (52). When using CT, tumor heterogeneity has been found to be a consistently useful feature, with metastases demonstrating increased heterogeneity compared with adenomas (24,45). In a study examining the apparent diffusion coefficient histogram analysis in adrenal nodules, heterogeneity was found to be significantly lower for adenomas compared with pheochromocytomas (48).

Our study specifically evaluated a common clinical problem, which is the management of an incidentally discovered adrenal nodule in a patient with cancer that does not show microscopic fat at chemical shift MRI. In the majority of patients, incidental adrenal nodules are almost always benign adenomas; however, the risk of malignancy increases in patients with cancer. It is reported widely that 30% of adenomas will be lipid poor (1,16) and in this setting, further confirmatory testing with adrenal washout CT or PET or through biopsy may be required. The actual rate of lipid-poor adenomas at MRI is unknown but expected to be less than the 30% figure cited, as a proportion of 10–30-HU adenomas can show microscopic fat at MRI. This, in combination with our stringent diagnostic criteria, likely accounted for the disparately higher number of metastases compared with lipid-poor adenomas identified during our study period. Although characterization of lipid-poor adrenal adenomas using adrenal washout CT is well documented with high reported accuracy (19,53,54), washout parameters are limited when assessing hypervascular metastases (8,52). Similarly, there is a reported overlap in fluorodeoxyglocose uptake in benign adenomas and metastases at PET (55–57). Adrenal biopsy to differentiate metastases from adenomas is generally diagnostic when performed adequately, but is invasive and may lead to increased risk of complications (58,59). Our results indicate that the use of T2-weighted MRI may accurately differentiate metastases from lipid-poor adenomas, potentially obviating additional testing in many patients. For example, subjectively assessed T2-weighted SI scores of 2 or less and heterogeneity scores of 3 or less were highly specific for the diagnosis of metastatic disease.

Our study had some limitations. We identified lipid-poor adenomas and metastases in patients who underwent MRI by using a keyword search. It is possible that despite consecutively retrieving eligible patients from our search, our sample is nonconsecutive if reports did not include those key terms; however, this risk should be balanced into both arms of the study. Including broader search terms such as “adrenal nodule” could have lessened but not eliminated this risk of bias. Our sample size was determined a priori, but there were fewer lipid-poor adenomas compared with metastases. The discrepant prevalence of lipid-poor adenomas compared with metastases in our study was almost certainly due to the stringent criteria used to establish a diagnosis of lipid-poor adenoma which excluded many nodules but was favored for specificity of diagnosis. Future studies including a larger number of lipid-poor adenomas could evaluate for potential improvements in accuracy by adding additional variables (including size and demographic variables) into logistic regression models. We also trained and tested our logistic regression models using the same data set, and this may overestimate accuracy, which would also be addressed with a future validation study. We did not include pheochromocytoma in the control group, because differentiation of pheochromocytoma from lipid-poor adenoma or metastases should be performed primarily based on biochemical testing and not by MRI (8). In terms of our reference standard, most samples were not histologically confirmed, but instead established by previously described imaging criteria, as well as stability, interval development, and greater than 20% interval growth on follow-up imaging studies. Metastases were larger than adenomas; therefore, we performed a subgroup analysis of nodules measuring less than 4 cm in size to validate our results. Future higher-powered studies might incorporate size as another variable in addition to T2-weighted SI and heterogeneity to determine if there is any incremental benefit on diagnostic accuracy. Another limitation in our study was the variation in MRI techniques, with a mixture of examinations performed at 1.5 T and 3 T. We used single-shot T2-weighted MRI for quantitative assessment for several reasons outlined above, but also because parameters such as matrix size and slice thickness, as well as echo time, are consistent between systems and vendors. These factors are known to cause variation in quantitative MRI analyses (21,40–43). We performed subgroup analysis of the data at 1.5 T and 3 T to ensure results were reproducible at low and high field strength. In addition, inherent noise reduction and surface coil correction algorithms may affect quantitative SI and histogram analysis. Two-dimensional single-slice technique was used, which improved practical applicability and reproducibility. The contribution of 3D whole-lesion analysis will require further assessment.

In conclusion, our study demonstrates that the combination of T2-weighted SI and T2-weighted heterogeneity, assessed subjectively and quantitatively (through T2-weighted entropy), was highly accurate to differentiate adrenal metastases from benign lipid-poor adrenal adenomas. Our results may enable the accurate characterization of incidentally discovered adrenal nodules that do not demonstrate microscopic fat at chemical shift MRI, potentially obviating further confirmatory imaging or biopsy. Future multicenter studies are required to validate these observations and propose thresholds, which could enable diagnosis in clinical practice.