Risk Stratification of Adrenal Masses by [18F]FDG PET/CT

Changing Tactics

Betty Salgues; Carole Guerin; Vincent Amodru; François Pattou; Laurent Brunaud; Jean-Christophe Lifante; Eric Mirallié; Nicolas Sahakian; Frédéric Castinetti; Anderson Loundou; Karine Baumstarck; Fréderic Sebag; David Taïeb


Clin Endocrinol. 2021;94(2):133-140. 

In This Article

Materials and Methods

Study Design

The inclusion criteria for validation cohort were (all criteria): 1—[18F]FDG-PET/CT performed between January 2017 and December 2019 at La Timone university hospital; 2—18F-[18F]FDG-PET/CT performed for characterization of an adrenal mass of maximal diameter ≥ 30 mm on axial CT or ≥ 20 mm and atypical features on adrenal CT (spontaneous density ≥ 10 HU and slow contrast washout (absolute washout < 60% and/or relative washout < 40%); 3—absence of previous history of any type of cancer, except remission > 5 yrs; and 4—Available complete hormonal work-up. Cases with elevated metanephrines were excluded.

The study was approved by the local ethical committee of Aix-Marseille University. All patients gave informed consent for the use of anonymous personal data extracted from their medical records for research purposes.

Patients and Tumour Secretory Status

Depending on their secretory status, patients were divided in four distinct categories: (a) Cushing syndrome when patients exhibited an absence of cortisol suppression (>50 nmol/L) after a low-dose dexamethasone suppression test, associated with suppressed ACTH secretion and the existence of comorbidities usually associated with hypercortisolism; (b) Subclinical hypercortisolism was defined following the ESE/ENSAT guidelines, as being likely when the cortisol after 1 mg-overnight dexamethasone suppression test was between 51 and 138 nmol/L (1.9–5.0 μg/dL) and certain when superior to 138 nmol/L, associated with suppressed ACTH secretion and the absence of comorbidities usually associated with hypercortisolism; (c) Other secretion defined by an increased of aldosterone/renin ratio or testosterone levels in women; (d) Nonsecreting, defined by normal urine and/or plasma metanephrines, normal aldosterone/renin ratio, normal mean of 2 measurements of 24-hour urinary free cortisol levels and cortisol level <50 nmol/L following a 1 mg-overnight dexamethasone suppression test, normal testosterone in women.


[18F]FDG-PET/CT was performed on a GE Healthcare Discovery PET/CT 710 (General Electric Healthcare) with the three-dimensional Time-Of-Flight mode. All patients fasted for 6 hours prior to scanning. [18F]FDG (3 MBq/kg) was intravenously injected. After tracer injection, patients remained at rest and [18F]FDG-PET/CT was acquired at approximately 1 hour post-injection. A whole-body imaging was performed from skull base to mid-thigh, corresponding to 6–8 steps of 2 minutes each. Slice thickness of the helical CT was 2.5 mm. The attenuation and impulsional response corrected PET was reconstructed with 3D iterative process (with 24 subsets and 2 iterations), using a CT attenuation map.

[18F]FDG-PET/CT scans were analysed by a nuclear physician investigator blinded to the results of the other imaging and biochemical studies.

Some parameters were extracted from CT attenuation corrected PET images:

  • Quantitative analysis of adrenal mass and contralateral adrenal gland uptake were assessed semi-automatically and expressed as SUVmax and tumour-to-liver SUVmax uptake ratio;

  • Visual analysis of the brown adipose tissue was performed;

  • Other parameters were extracted from unenhanced CT images:

  • Tumour size and density;

  • Description of the contralateral adrenal gland as follows: 0—normal, 1—hyperplasia and 2—nodular;

  • Analysis of liver steatosis;

  • Measurements of total (T), visceral (V), subcutaneous (S) fat and muscle areas as follows:

The scan being measured was loaded into the Volume Viewer to enable the L3-L4 disc space to be identified using sagittal views.

The single cross-sectional CT image (L3-L4 disc space) was saved as a DICOM and imported into CoreSlicer ( https://coreslicer.com ).[11] Attenuation range of HU was set to provide tissue areas (muscle, adipose tissue). Range of HU to select pixels was unknown by users because of the proposed workflow. On the interface, the tissue area is directly selected (muscle, subcutaneous or visceral adipose tissue, bone) and not range of HU. Automated segmentation algorithms are referenced and used in this study to provide body composition. Manual corrections were performed to delete aberrant pixel or add others in the body segmentation. Analysis of fat and lean tissue at this area is highly correlated to corresponding whole-body composition and is an important predictor of the metabolic syndrome.[12–16] Total body fat mass and lean body mass have been defined using equations published by Mourtzakis et al.[13]

Gold standard

Gold standard was defined from:

  • Histology: (a) for adrenocortical tumours, malignant tumours was based on a Weiss score for ACCs ≥ 3; (b) for oncocytomas, malignant was based on Bisceglia scoring system: the existence of at least one major criterion defines a malignant oncocytoma. Oncocytomas with uncertain malignant potential (borderline) defined by one to four minor criteria were classified as benign.

  • For nonoperated adrenal masses: lesions were classified as benign by a multidisciplinary staff based on imaging features on CT and CT follow-up (≥6 months). Stable disease was assumed when the mass remained stable or had minimal increase in size (<15%) of the tumour diameter on the last CT.

Statistical Analysis

Statistical analysis was performed using IBM SPSS Statistics version 20 (IBM SPSS Inc). Continuous variables are expressed as means ± SD or medians with range (min, max), and categorical variables are reported as count and percentages. All the tests were two-sided. Statistical significance was defined as P < .05.

Two different populations were defined: all the cases and the subgroup of adrenocortical tumours. For each population, comparisons of imaging findings between two groups, benign and malignant masses, were performed using Student t test or Mann-Whitney U for continuous variables and chi-Square test (or Fisher's exact test, as appropriate) for qualitative variables.

First, a prediction model was built from a prospective cohort (training cohort) using a logistic regression model including 2 parameters: the tumour SUVmax: liver SUVmax uptake ratio and the tumour diameter. To quantify the discrimination performance of the model, the area under the receiver operating characteristic (AUC) curve was measured. Calibration plots were used to assess the calibration of this model, accompanied with a Hosmer-Lemeshow.

The prospective training cohort consisted of 87 patients from 8 French university hospitals and has been previously described.[10] In the training cohort, among the 87 masses, 72 were classified as benign and 15 as malignant. Briefly, histology was obtained in 64 patients and identified 15 malignant tumours (11 ACCs, one metastasis from lung carcinoma, two leiomyosarcomas, one liposarcoma); 47 benign tumours including 34 ACAs and two oncocytomas with uncertain malignant potential. The remaining 23 cases remained stable on 12 months follow-up CT and were therefore considered as benign lesions.

Second, the performance of the model was tested on an independent cohort (validation cohort). The logistic regression formula from the initial cohort was applied to the validation cohort, and the probability for each patient was calculated. To quantify the discrimination performance of the model, the AUC was measured. Calibration plots were used to assess the calibration of this model, accompanied with a Hosmer-Lemeshow chi-square test. The cut-off points were calculated from the ROC curves that maximized both sensitivity (Se) and specificity (Sp). NPV, PPV and accuracy were provided with their 95% confidence intervals.