Radiomics Analysis of [ 18 F]-Fluorodeoxyglucose-Avid Thyroid Incidentalomas Improves Risk Stratification and Selection for Clinical Assessment

Abstract

Background:

[¹⁸F]-Fluorodeoxyglucose (FDG)-avid thyroid lesions incidentally detected on positron emission tomography/computed tomography (PET/CT) scans represent a tumor lesion in about 30% of cases. The present study evaluated the ability of PET metrics and radiomics features to predict final diagnosis of [¹⁸F]FDG thyroid incidentalomas (TIs).

Methods:

A total of 104 patients with 107 TIs were retrospectively studied; 30 nodules (28%) were diagnosed as malignant. After volumetric segmentation of each thyroid lesion, metabolic tumor volume (MTV), total lesion glycolysis (TLG), standardized uptake values (SUVs), and metabolic heterogeneity were estimated, and 107 radiomics features were extracted following a standard protocol.

Results:

MTV, TLG, SUVmax, SUVmean, and SUVpeak among functional PET parameters, and gray-level co-occurrence matrix (GLCM)_InverseDifferenceMoment, shape_Sphericity, GLCM_SumSquares, firstorder_Maximum2DDiameterSlice, firstorder_Energy, and GLCM_Contrast among nonredundant radiomics features, showed significantly different values between malignant and benign TIs (Mann–Whitney U-test, p < 0.01 for all). Univariate logistic regression revealed that these parameters demonstrated good ability to predict final diagnosis of TIs (p < 0.02 for all). Shape_Sphericity was the best predictor classifying 82% of TIs correctly (p < 0.0001). Only TLG, SUVmax, and shape_Sphericity retained significance (p < 0.0001) by multivariate analysis. Malignant lesion prevalence increased from 7% to 100% in accordance with the number (score, 0–3) of the three positive parameters present (χ² trend, p < 0.0001). A score of 0 excludes malignant TIs with a negative predictive value of 93%, while a score of 3 predicted malignancy with a positive predictive value of 100%.

Conclusions:

PET metrics and radiomics analysis can improve identification of [¹⁸F]FDG-avid TIs at high risk of malignancy. A model based on TLG, SUVmax, and shape_Sphericity may allow prediction of a final diagnosis, providing useful information for the management of TIs.

Introduction

Incidentally discovered [¹⁸F]-fluorodeoxyglucose (FDG)-avid thyroid incidentalomas (TIs) are reported in about 1.5–2% of all positron emission tomography/computed tomography (PET/CT) scans performed for nonthyroid diseases (1 –3), and about 30% are malignant (1,2,4,5). Unfortunately, neither visual analysis nor PET-derived parameters, such as standardized uptake values (SUVs), metabolic tumor volume (MTV), and total lesion glycolysis (TLG), reliably discriminate malignant from benign [¹⁸F]FDG-avid TIs (3,6 –11). Ultrasound (US) assessment is pivotal in detection and exclusion of thyroid cancers among patients with thyroid nodules, and recently, several US risk stratification systems have been used in clinical practice (12). However, studies focusing only on the performance of US have provided sparse results (13 –15), and limited preliminary data are currently available on US risk stratification systems used in patients with [¹⁸F]FDG-TIs (16,17).

International clinical guidelines strongly recommend use of fine-needle aspiration cytology to ascertain [¹⁸F]FDG-avid TIs greater than 1 cm in greatest diameter (5). However, the prognosis of oncology patients is dictated by the underlying malignancy rather than by incidentally detected thyroid cancers in most cases. Accordingly, the costs, risks, and anxiety associated with incidentally detected [¹⁸F]FDG-avid thyroid carcinomas should be carefully balanced by primary malignancy (18 –20).

Accurate PET-derived criteria would be helpful to directly stratify the malignancy risk of [¹⁸F]FDG-TIs during PET/CT interpretation and reduce the need for further evaluation and the associated risks, costs, and patient discomfort. Recent preliminary data have suggested that heterogeneous [¹⁸F]FDG uptake within thyroid nodules (i.e., “metabolic heterogeneity”) and PET/CT texture analysis are promising approaches to stratify the risk of a malignant nodule in patients with [¹⁸F]FDG-avid incidentalomas, refine selection of patients for cytology referral, and reduce the need for further investigations (21 –23). The present study investigated the role of PET-derived parameters and radiomics features to characterize [¹⁸F]FDG-TIs in a large series of patients with cancer and generated a predictive model for discriminating malignant from benign [¹⁸F]FDG-TIs based on PET/CT-derived data.

Materials and Methods

Study design and patient selection

Initially, our facility's database was searched for all patients who underwent [¹⁸F]FDG-PET/CT due to nonthyroid oncological indications from October 2013 to November 2018. Then, all cases with abnormal thyroid [¹⁸F]FDG uptake were retrieved. After exclusion of cases with diffuse [¹⁸F]FDG thyroid uptake, patients with focal [¹⁸F]FDG uptake were selected. The study inclusion criteria were as follows: (i) a histological diagnosis, (ii) benign cytological report in the absence of postoperative histology, and (iii) scintigraphically confirmed autonomously functioning thyroid nodules that received curative treatment with different modalities (24). All fine-needle aspiration cytology was reported according to the Italian Consensus for the Classification and Reporting of Thyroid Cytology [i.e., TIR1 (inadequate), TIR2 (benign), TIR3 (indeterminate), TIR4 (suspicious for malignancy), and TIR5 (malignancy)] (25).

[¹⁸F]FDG-PET/CT image acquisition

[¹⁸F]FDG-PET/CT whole-body images were acquired following a standard protocol in accordance with European Association of Nuclear Medicine guidelines (26) using two integrated PET/CT scanners of the same model (both Siemens Biograph mCT 40; Siemens, Erlangen, Germany). Scanning was performed 60 ± 5 minutes after intravenous injection of 3 MBq/kg of [¹⁸F]FDG. All patients fasted for at least 6 hours before image acquisition and their fasting blood glucose was <160 mg/dL. CT scans obtained with a low-dose protocol were used for attenuation-correction of PET images. Images were reconstructed with three-dimensional ordered subset expectation maximization iterative algorithms (3 iterations and 21 subsets), including Time-of-Flight and Point-Spread Function corrections.

[¹⁸F]FDG-PET/CT metrics

Reconstructed [¹⁸F]FDG-PET/CT images were analyzed with dedicated software (MM Oncology, Syngo.via; Siemens). An expert nuclear medicine physician (L.C.) contoured all thyroid focal PET findings with a three-dimensional fixed threshold algorithm. The MTV of each TI was automatically estimated setting the mean SUV of the contralateral lobe as threshold to identify the borders between pathological and normal tissue. Then, maximum, mean, and peak SUVs (SUVmax, SUVmean, and SUVpeak, respectively) of the lesion were measured automatically. The TLG was calculated as the product of the SUVmean and MTV (27). Metabolic heterogeneity was assessed using the cumulative SUV volume histograms (CSH) method (28), as applied in previous studies (29 –34). The area under the curve of the CSH is a quantitative index of tracer uptake heterogeneity where lower values correspond to increased heterogeneity (35). Metabolic heterogeneity was also evaluated by estimating the coefficient of variation of intra-lesion [¹⁸F]FDG uptake, which was calculated as the standard deviation of the SUVs divided by the SUVmean of the segmented lesion (30,36).

Radiomics features

Radiomics features were extracted from each segmented volume using PyRadiomics software package version 2.2.0 (37). To standardize the process of extraction, gray-level intensities and voxel dimensions of the original images were preliminary resampled following the Image Biomarkers Standardization Initiative recommendations (38 –41). From the segmented volumes, 107 standardized features evaluating different metabolic characteristics of the lesion were initially extracted. They included 14 shape-based features, 18 first-order statistics features, and 75 matrix-based features [24 gray-level co-occurrence matrix (GLCM), 16 gray-level run length matrix, 16 gray-level size zone matrix, 5 neighboring gray tone difference matrix, and 14 gray-level dependence matrix based]. Meaning and mathematical description of these radiomics features are reported in detail in the PyRadiomics documentation.

The shape-based features measure the dimensions and depict the geometric properties of the lesion. The first-order statistics features describe the presence of areas with statistically different metabolic activity without taking into account their distribution within the lesion, while the matrix-based features analyze spatial distribution of these areas and their mutual relationship. In the last group, the analysis is performed applying multiple mathematical models to the data organized in the form of matrices.

Since the image was acquired with two different scanners, the first analytic step was aimed at evaluating the inter-scanner reproducibility of the radiomics features. For each feature, the distributions of the values extracted by the two sets of images were compared. The features with different distributions were excluded from the final analysis to avoid a potential bias, while those with statistically proved inter-scanner reproducibility were tested as potential predictors of the final diagnosis of TIs.

Reference standard

Malignant lesions described throughout the article were thyroid cancers confirmed by surgical pathology examination. Benign lesions were defined as follows: histologically confirmed benign nodules (42), cytologically benign nodules (25) with clinical and US follow-up ≥3 years, and scintigraphically confirmed autonomously functioning thyroid nodules (43).

Ethics

This retrospective study on existing patient data and images was reviewed and approved by our Advisory Research Board and the Ethic Committee of Canton Tessin, Switzerland (ref. BASEC 2018-00685, CE TI 3351). The requirement for informed consent was waived. All procedures were in accordance with the ethical standards of the 1964 Declaration of Helsinki and its later amendments.

Statistical analyses

Quantitative variables are expressed as medians and interquartile ranges. Medians were compared by the Mann–Whitney U-test. Differences between categorical data frequency were assessed by χ² test. Reproducible radiomics features were identified by comparing results from the two PET-CT scanners, applying the Mann–Whitney U-test with false discovery rate correction. Nonredundant and uncorrelated radiomics features were selected using the correlation-based feature selection algorithm, a supervised method for reducing the dimensionality. Continuous parameters were analyzed as dichotomized variables using receiver-operating characteristic curves and Youden's coefficient method to estimate the optimal cutoff point for discriminating malignant from benign TIs. Univariate logistic regression analysis was applied to assess the relationship between each parameter and the final diagnosis. Logistic stepwise regression function was used for multivariate analysis. To evaluate the robustness and the generalization of our model and considering the sample size of the study, a 1000-resampled bootstrapping was performed as cross-validation procedure. A p < 0.05 was considered statistically significant. Negative and positive predictive values were calculated. Statistical analyses were conducted using RStudio statistical software package version 1.2.503 (RStudio, Inc., Boston, MA) and MedCalc Statistical Software version 15.8 (MedCalc Software bvba, Ostend, Belgium).

Results

Characteristics of patients

Among the 12,652 consecutive patients who underwent [¹⁸F]FDG-PET/CT during the study period, 333 (2.6%) TIs were recorded. After exclusion of cases with diffuse [¹⁸F]FDG thyroid uptake patterns (n = 146), 187 (57%) [¹⁸F]FDG-TIs were included. According to the selection criteria, 107 focal TIs from 104 patients (67 females and 37 males; median age, 65 years) were selected for the present analysis (Fig. 1). The median thyrotropin was 1.5 mIU/L (0.96–2.1 mIU/L). Lesions had a median size of 17 mm (12–26 mm). As summarized in Table 1, 30/107 (28%) TIs were malignant (26 primary thyroid carcinomas, 1 non-Hodgkin lymphoma, and 3 metastatic lesions from sarcoma, renal, and esophageal cancer, respectively) and 77 (72%) were benign (14 autoimmune thyroiditis, 1 autonomously functioning thyroid nodules, and 62 hyperplastic nodules).

FIG. 1.

Patient evaluations and results of diagnostic workup. ATC, anaplastic thyroid carcinoma; FNAC, fine-needle aspiration cytology; FTC, follicular thyroid carcinoma; MTC, medullary thyroid carcinoma; PTC, papillary thyroid carcinoma; TIR, FNAC results reported according to the Italian Consensus for the Classification and Reporting of Thyroid Cytology [TIR1 (inadequate), TIR2 (benign), TIR3 (indeterminate), TIR4 (suspicious for malignancy), TIR5 (malignancy)]; US FU: ultrasounds follow-up.

Table 1.

Demographic and Clinicopathologic Characteristics of Patients

Variable	Patients (n = 104)	Benign lesions (n = 77)	Malignant lesions (n = 30)	p^***
Sex (female/male)^*	67/37 (64%/36%)	51/26 (66%/34%)	19/11 (63%/37%)	0.95
Age (years)^**	65 (54–74)	65 (54–75)	62.5 (52–72)	0.27
TSH (mUI/L)^**	1.5 (0.96–2.10)	1.5 (0.86–2.10)	1.6 (1.2–2.80)	0.15
Largest tumor diameter on US (mm)^**	17 (12 –26)	17 (12–23.5)	21 (14–30.5)	0.10
Final diagnosis		Benign goiter 62	Thyroid carcinoma 26
		AFTN 1	Papillary 19
		AIT 14	Follicular 5
			Medullary 1
			Anaplastic 1
			Non-Hodgkin lymphoma 1
			Metastatic lesion 3
			Sarcoma 1
			Renal cancer 1
			Esophageal cancer 1

Data are expressed as ^*frequency or ^**median and interquartile range.

***

χ² test for comparison of frequencies, and the Mann–Whitney U-test for comparison of medians.

AFTN, autonomously functioning thyroid nodules; AIT, autoimmune thyroiditis; TSH, thyrotropin; US, ultrasound.

PET metrics and radiomics analysis

Among functional and volumetric PET-derived parameters, MTV, TLG, SUVmax, SUVmean, and SUVpeak values were significantly increased in malignant compared with benign TIs (Table 2). After dichotomization, such parameters reliably predicted the final diagnosis on univariate analysis. TLG was the best predictor, correctly classifying 79% of lesions (univariate logistic regression, p < 0.0001). Specifically, low TLG values were associated with benign lesions and had a negative predictive value of 84%, while higher TLG values had a positive predictive value of 65% for malignant nodules. However, metabolic heterogeneity assessed either by area under the curve-CSH or by coefficient of variation was not associated with the final diagnosis due to the wide overlap of values between the two groups.

Table 2.

Distribution of Positron Emission Tomography Functional Parameters and Nonredundant Radiomics Features in the Two Subgroups of Patients with Benign and Malignant Nodular Disease

Variable	Benign lesions (n = 77)	Malignant lesions (n = 30)	p^*
PET metrics
SUVmax	6.71 (4.52–10.96)	10.81 (7.21–14.71)	0.0146
SUVmean	2.79 (2.39–3.57)	3.88 (2.78–5.64)	0.0040
SUVpeak	3.45 (2.98–5.57)	7.11 (4.25–10.90)	0.0005
MTV (mL)	3.64 (2.48–7.22)	10.75 (4.10–29.66)	0.0002
TLG	11.61 (8.73–15.30)	44.69 (19.55–81.28)	<0.0001
AUC-CSH	0.42 (0.40–0.50)	0.44 (0.42–0.50)	0.5930
CV	0.40 (0.29–0.59)	0.54 (0.28–0.65)	0.2580
Radiomics
Shape_Sphericity	0.67 (0.53–0.78)	0.79 (0.72–0.82)	0.0004
Shape_Maximum2DDiameterSlice	32.20 (23.20–45.30)	22.85 (20.12–29.55)	0.0018
Firstorder_Energy	5592.0 (1533.40–20203.00)	1178 (629.30–3541.40)	0.0005
GLCM_Contrast	150.30 (86.50–265.20)	253.8 (179.10–328.10)	0.0025
GLCM_InverseDifferenceMoment	0.13 (0.11–0.18)	0.09 (0.07–0.13)	0.0003
GLCM_SumSquares	183.20 (142.30–226.10)	212.10 (185.80–239.30)	0.0380

Data are expressed as median and interquartile range.

Mann–Whitney U-test.

AUC-CSH, area under the curve-cumulative SUV volume histograms; CV, coefficient of variation; GLCM, gray-level co-occurrence matrix; MTV, metabolic tumor volume; PET, positron emission tomography; SUV, standardized uptake value; TLG, total lesion glycolysis.

Among the 107 extracted radiomics features, only 54 were statistically reproducible between the two PET/CT scanners. Among these, the correlation-based feature selection algorithm selected six nonredundant and uncorrelated radiomics features as potential predictors of the final diagnosis. The six selected features are summarized in Table 2. The shape_Sphericity and the shape_Maximum2DDiameterSlice are two morphological features measuring the degree of spherical shape and the largest dimension in the axial plane of the lesion, respectively.

The firstorder_Energy is a gray-level histogram-based feature measuring the intensity of voxel values (i.e., the metabolic activity) in the image. The GLCM_Contrast, GLCM_InverseDifferenceMoment, and GLCM_SumSquares are three textural features belonging to the group of the GLCM-based features that describe with different tools the degree of metabolic heterogeneity of the TI.

Their ability to distinguish malignant from benign TIs was significant by the Mann–Whitney U-test (Table 2) and univariate logistic regression of dichotomized data (Table 3). Shape_Sphericity had the highest accuracy, correctly classifying 82% of TIs (univariate logistic regression, p < 0.0001). TIs closer to the spherical shape (i.e., with higher values of sphericity) had a lower risk of malignancy with negative predictive values of 82%, while the lack of sphericity identified a tumor lesion with positive predictive value of 82% (Table 3). In multivariate stepwise logistic regression analysis including the previously selected five functional PET parameters and six radiomics features, TLG, SUVmax, and shape_Sphericity retained statistical significance (p < 0.0001; Table 4). The 1000-resampled bootstrapping method validated the multivariate analysis results with an optimism-corrected area under the curve of 0.830 (p < 0.0001).

Table 3.

Positron Emission Tomography Metrics and Radiomics Parameters, Receiver Operator Characteristic Curve Analysis and Univariate Logistic Regression Results

Variable	AUC	Best cut-point	p	Sensitivity (%)	Specificity (%)	Logistic regression p-value	OR [CI]	NPV (%)	PPV (%)	Accuracy (%)
SUVmax	0.652	>7.05	0.0133	76.7	57.1	0.0025	4.38 [1.68–11.43]	86	41	74
SUVmean	0.68	>3.03	0.0037	50.0	83.1	0.0008	4.92 [1.94–12.5]	81	54	74
SUVpeak	0.717	>3.91	0.0003	80.0	58.4	0.0007	5.62 [2.06–15.33]	88	43	64
MTV	0.733	>8.63	0.0001	60.0	81.8	0.0003	5.29 [2.14–13.12]	83	51	73
TLG	0.756	>39.15	<0.0001	56.7	88.3	<0.0001	9.88 [3.63–26.92]	84	65	79
AUC-CSH	0.533	>0.42	0.580	66.7	50.6	0.1096	2.05 [0.85–4.95]	80	34	55
CV	0.571	>0.59	0.283	43.3	76.6	0.0441	2.51 [1.02–6.13]	78	42	67
Shape_Sphericity	0.723	<57.99	0.0004	46.7	96.1	<0.0001	21.58 [5.54–84.01]	82	82	82
Shape_Maximum2D DiameterSlice	0.695	>31.32	0.0018	53.3	84.4	0.0005	5.42 [2.10–13.92]	81	55	75
Firstorder_Energy	0.716	>1177.97	0.0003	86.7	50.6	0.0011	6.67 [2.13–20.93]	91	41	61
GLCM_Contrast	0.699	<163.53	0.0013	56.7	84.4	0.0002	6.19 [2.40–15.93]	82	57	76
GLCM_InverseDifferenceMoment	0.728	>10	<0.0001	83.3	55.8	0.0007	6.32 [2.19–18.25]	89	42	64
GLCM_SumSquares	0.629	<177.59	0.0526	46.7	83.1	0.0021	4.31 [1.69–10.94]	80	52	73

CI, 95% confidence interval; NPV, negative predictive value; OR, odds ratio; PPV, positive predictive value.

Table 4.

Multivariate Analysis (Stepwise Logistic Regression) Results

Variable	OR [CI]	p	Likelihood ratio test	Hosmer and Lemeshow goodness of fit test	Bootstrap (R = 1000) optimism-corrected AUC-ROC
SUVmax	4.15 [1.22–14.19]	0.0231	<0.0001	0.9999	0.830
TLG	4.52 [1.39–14.71]	0.0123
Shape_Sphericity	18.11 [3.68–89.11]	0.0004

Results including the five functional PET parameters and six radiomics features that were significant at univariate analysis.

AUC-ROC, area under the curve-receiver operator characteristic.

Multiparametric predictive model

Based on these results, we tested a predictive model integrating the dichotomized TLG, SUVmax, and shape_Sphericity. In this model, four subgroups of lesions were defined by a score from 0 to 3 reproducing the number of positive imaging-derived parameters. This model was accurate for stratifying the risk of TI malignancy (χ² test, p < 0.0001). The prevalence of malignant TIs increased from 7% (3/41) for lesions with a score of 0 to 21% (9/42), 60% (9/15), and 100% (9/9) for lesions with scores of 1, 2, and 3, respectively (χ² trend, p < 0.0001). Risk of malignancy only remained comparable to that of the overall population in the group with a score of 1 (χ² test p = 0.503), while the group with a score of 0 was significantly lower (χ² test, p = 0.0114) and groups with scores of 2 and 3 were significantly higher (χ² test, p = 0.0282 and <0.0001, respectively). In summary, a score of 0 identified benign TIs with a negative predictive value of 93%, while scores of 2 and 3 predicted malignancy with positive predictive values of 60% and 100%, respectively.

Discussion

Assessment of [¹⁸F]FDG TIs is a critical challenge for oncology patients as the reported thyroid cancer prevalence ranges from 20% to 40%. Furthermore, patients undergoing [¹⁸F]FDG-PET/CT are generally affected by aggressive nonthyroid cancers that commonly dictate their prognosis. Thus, developing a PET/CT-based strategy to stratify the risk of malignancy in FDG-TIs may assist nuclear medicine physicians during PET/CT reporting and provide patient-specific risk stratification. In turn, additional referrals and investigations could be reduced, positively impacting patient quality of life and reducing overall costs.

Previous studies have reported increased SUVmax and SUVmean values and higher PET volume-based measurements (MTV and TLG) in malignant TIs, suggesting a potential role for these parameters, alone or in combination, as predictors of TI malignancy (7,9,10,22). Others, however, have reported no significant association of the same parameters with TI malignancy (1,6,8,11). More recently, a different approach based on assessment of [¹⁸F]FDG distribution within thyroid nodules has been proposed (21 –23), and preliminary data suggest that high metabolic heterogeneity may identify malignant TIs better than conventional PET metrics (21). Sollini et al. reported interesting preliminary results by evaluating histogram-based and matrix-based features by textural analysis (22,23).

To the best of our knowledge, our study is the first attempt to integrate textural features and conventional PET metrics in a multiparametric predictive model for TIs. The present data confirmed the predictive power of conventional PET metrics parameters and identified six independent radiomics features describing shape, heterogeneity, and intensity of lesion tracer uptake as additional tools for discriminating benign from malignant TIs. As the main result, a predictive model combining the three independent PET-derived predictive parameters (i.e., TLG, SUVmax, and shape_Sphericity) proved to be highly accurate for stratifying the risk of malignancy of [¹⁸F]FDG-avid TIs. In particular, all triple-positive TIs were malignant (positive predictive value, 100%), while 93% of triple-negative cases were benign.

The present retrospective single-center study has some limitations. First, in some cases, the severity of the primary malignancy could have prevented a complete diagnostic workup of the TI and the exclusion of these patient in our analysis. However, the prevalence of thyroid cancer among the study cohort was in line with current rates reported in the literature. Second, a validated threshold value to segment thyroid PET-positive thyroid nodules has not been defined yet. However, our approach was based on arbitrarily selecting the SUVmean of the contralateral lobe to define the actual volume of TIs independent of their metabolic activity increasing accuracy and reproducibility of radiomics analysis. Third, we analyzed only lesions larger than 1.1 cm, consistent with the current recommendation to not perform fine-needle aspiration cytology in nodules, included FDG-active ones, less than 1 cm in largest diameters (5). Thus, our findings may not be applicable to smaller TIs. Fourth, even if the present series is larger than those previously reported, validation of the current model in a prospective study including a larger number of cases is warranted to confirm these results.

Finally, we did not compare PET data with US pattern in our series as US examinations were performed by different sonographers in different centers precluding any reliable comparison. However, it should be noted that, even if the potential role of US in this setting was reported (17), current clinical guidelines still recommend fine-needle aspiration cytology in all FDG-active TIs >1 cm (5). Accordingly, the aim of this study was to test the possibility of better characterizing metabolic PET information to guide the need for fine-needle aspiration in patients with cancer independent of other evaluation. Our PET/CT-based model was able to predict the risk of malignancy of PET TIs. Such approach may potentially avoid unnecessary additional procedures. The potential improvement generated by the integration between PET and US risk factors remains of course of high interest and deserves to be explored in future studies.

In conclusion, the present multiparametric PET/CT-based radiomics model showed good performance in stratifying the risk of malignancy of [¹⁸F]FDG-avid TIs and may be useful for reducing the number of fine-needle aspiration cytology referrals in patients with cancer.

Footnotes

Author Disclosure Statement

The authors have no competing financial interests to declare.

Funding Information

The authors received no financial support for the study.

References

Bertagna

, Treglia

, Piccardo

, Giubbini

. 2012. Diagnostic and clinical significance of F-18-FDG-PET/CT thyroid incidentalomas. J Clin Endocrinol Metab, 97:3866–3875.

Treglia

, Giovanella

, Bertagna

, Di Franco

, Salvatori

. 2013. Prevalence and risk of malignancy of thyroid incidentalomas detected by (18)f-fluorodeoxyglucose positron-emission tomography. Thyroid, 23:124–126.

Signore

, Albano

, Giovanella

, Bertagna

, Treglia

. 2020. Evidence-based data about prevalence and risk of malignancy of thyroid incidentalomas detected by different PET radiopharmaceuticals. Curr Radiopharm, 13:89–93.

Treglia

, Bertagna

, Sadeghi

, Verburg

, Ceriani

, Giovanella

. 2013. Focal thyroid incidental uptake detected by (1)(8)F-fluorodeoxyglucose positron emission tomography. Meta-analysis on prevalence and malignancy risk. Nuklearmedizin, 52:130–136.

Haugen

, Alexander

, Bible

, Doherty

, Mandel

, Nikiforov

, Pacini

, Randolph

, Sawka

, Schlumberger

, Schuff

, Sherman

, Sosa

, Steward

, Tuttle

, Wartofsky

. 2016. 2015 American Thyroid Association Management Guidelines for Adult Patients with Thyroid Nodules and Differentiated Thyroid Cancer: the American Thyroid Association Guidelines Task Force on Thyroid Nodules and Differentiated Thyroid Cancer. Thyroid, 26:1–133.

Sager

, Vatankulu

, Sahin

, Cınaral

, Uslu

, Baran

, Ozturk

, Sönmezoglu

. 2018. Clinical significance of standardized uptake values in thyroid incidentaloma discovered by F-18 fluorodeoxyglucose positron emission tomography/computed tomography. J Cancer Res Ther, 14:989–993.

Shi

, Yuan

, Yang

, Zhang

, Shou

, Zhang

, Ping

, Gao

, Liu

. 2018. Diagnostic value of volume-based fluorine-18-fluorodeoxyglucose PET/CT parameters for characterizing thyroid incidentaloma. Korean J Radiol, 19:342–351.

Thuillier

, Bourhis

, Roudaut

, Crouzeix

, Alavi

, Schick

, Robin

, Kerlan

, Salaun

P-Y

, Abgral

. 2019. Diagnostic value of FDG PET-CT quantitative parameters and Deauville-like 5 point-scale in predicting malignancy of focal thyroid incidentaloma. Front Med (Lausanne), 6:24.

Kim

, Kim

S-J

, Kim

, Jeon

, Kim

. 2013. Diagnostic value of metabolic tumor volume assessed by 18F-FDG PET/CT added to SUVmax for characterization of thyroid 18F-FDG incidentaloma. Nucl Med Commun, 34:868–876.

10.

Soelberg

, Bonnema

, Brix

, Hegedüs

. 2012. Risk of malignancy in thyroid incidentalomas detected by 18F-fluorodeoxyglucose positron emission tomography: a systematic review. Thyroid, 22:918–925.

11.

T-Y

, Liou

M-J

, Lin

K-J

, Yen

T-C

. 2011. Prevalence and significance of thyroid uptake detected by ¹⁸F-FDG PET. Endocrine, 40:297–302.

12.

Castellana

, Castellana

, Treglia

, Giorgino

, Giovanella

, Russ

, Trimboli

. 2019. Performance of five ultrasound risk stratification systems in selecting thyroid nodules for FNA. A meta-analysis. J Clin Endocrinol Metab, 105:dgz170.

13.

Lin

, Tsai

, Lin

, Der Lin

, Wang

, Chen

. 2019. computer-aided diagnostic technique in 2-deoxy-2-[(18)F]fluoro-D-glucose-positive thyroid nodule: clinical experience of 74 non-thyroid cancer patients. Ultrasound Med Biol, 45:108–121.

14.

Chung

, Choi

, Suh

, Kim

, Lee

, Kim

, Sung

, Lee

, Song

, Lee

, Baek

. 2018. Thyroid incidentalomas detected on (18)F-fluorodeoxyglucose positron emission tomography with computed tomography: malignant risk stratification and management plan. Thyroid, 28:762–768.

15.

Beech

, Lavender

, Jong

, Soo

, Ramdave

, Chong

, Nandurkar

. 2016. Ultrasound stratification of the FDG-avid thyroid nodule. Clin Radiol, 71:164–169.

16.

Yoon

, Cho

, Lee

, Kim

, Moon

, Kwak

. 2015. Thyroid incidentalomas detected on 18F-fluorodeoxyglucose-positron emission tomography/computed tomography: thyroid Imaging Reporting and Data System (TIRADS) in the diagnosis and management of patients. Surgery, 158:1314–1322.

17.

Trimboli

, Paone

, Treglia

, Virili

, Ruberto

, Ceriani

, Piccardo

, Giovanella

. 2018. Fine-needle aspiration in all thyroid incidentalomas at (18) F-FDG PET/CT: can EU-TIRADS revise the dogma?. Clin Endocrinol, 89:642–648.

18.

Poller

, Bongiovanni

, Trimboli

. 2020. Risk of malignancy in the various categories of the UK Royal College of Pathologists Thy terminology for thyroid FNA cytology: a systematic review and meta-analysis. Cancer Cytopathol, 128:36–42.

19.

Trimboli

, Fulciniti

, Paone

, Barizzi

, Piccardo

, Merlo

, Mazzucchelli

, Giovanella

. 2020. Risk of malignancy (ROM) of thyroid FNA diagnosed as suspicious for malignancy or malignant: an institutional experience with systematic review and meta-analysis of literature. Endocr Pathol, 31:52–56.

20.

Pattison

, Bozin

, Gorelik

, Hofman

, Hicks

, Skandarajah

. 2018. (18)F-FDG-avid thyroid incidentalomas: the importance of contextual interpretation. J Nucl Med, 59:749–755.

21.

Kim

S-J

, Chang

. 2015. Predictive value of intratumoral heterogeneity of F-18 FDG uptake for characterization of thyroid nodules according to Bethesda categories of fine needle aspiration biopsy results. Endocrine, 50:681–688.

22.

Sollini

, Cozzi

, Pepe

, Antunovic

, Lania

, Di Tommaso

, Magnoni

, Erba

, Kirienko

. 2017. [(18)F]FDG-PET/CT texture analysis in thyroid incidentalomas: preliminary results. Eur J Hybrid Imaging, 1:3.

23.

Sollini

, Kirienko

, Magnoni

, Cozzi

, di Tommaso

, Lania

, Arturo

. 2019. [18F]FDG-PET/CT texture analysis in thyroid “hot” nodules: results of the validation cohort. Biomed J Sci Techn Res, 18:13344–13350.

24.

Giovanella

, Piccardo

, Pezzoli

, Bini

, Ricci

, Ruberto

, Trimboli

. 2018. Comparison of high intensity focused ultrasound and radioiodine for treating toxic thyroid nodules. Clin Endocrinol, 89:219–225.

25.

Nardi

, Basolo

, Crescenzi

, Fadda

, Frasoldati

, Orlandi

, Palombini

, Papini

, Zini

, Pontecorvi

, Vitti

. 2014. Italian consensus for the classification and reporting of thyroid cytology. J Endocrinol Invest, 37:593–599.

26.

Boellaard

, Delgado-Bolton

, Oyen

WJG

, Giammarile

, Tatsch

, Eschner

, Verzijlbergen

, Barrington

, Pike

, Weber

, Stroobants

, Delbeke

, Donohoe

, Holbrook

, Graham

, Testanera

, Hoekstra

, Zijlstra

, Visser

, Hoekstra

, Pruim

, Willemsen

, Arends

, Kotzerke

, Bockisch

, Beyer

, Chiti

, Krause

, European Association of Nuclear

. 2015. FDG PET/CT: EANM procedure guidelines for tumour imaging: version 2.0. Eur J Nucl Med Mol Imaging, 42:328–354.

27.

Larson

, Erdi

, Akhurst

, Mazumdar

, Macapinlac

, Finn

, Casilla

, Fazzari

, Srivastava

, Yeung

, Humm

, Guillem

, Downey

, Karpeh

, Cohen

, Ginsberg

. 1999. Tumor treatment response based on visual and quantitative changes in global tumor glycolysis using PET-FDG imaging. The visual response score and the change in total lesion glycolysis. Clin Positr Imaging, 2:159–171.

28.

El Naqa

, Grigsby

, Apte

, Kidd

, Donnelly

, Khullar

, Chaudhari

, Yang

, Schmitt

, Laforest

, Thorstad

, Deasy

. 2009. Exploring feature-based approaches in PET images for predicting cancer treatment outcomes. Pattern Recogn, 42:1162–1171.

29.

Yoon

, Kim

. 2015. Intratumoral metabolic heterogeneity predicts invasive components in breast ductal carcinoma in situ. Eur Radiol, 25:3648–3658.

30.

Watabe

, Tatsumi

, Watabe

, Isohashi

, Kato

, Yanagawa

, Shimosegawa

, Hatazawa

. 2012. Intratumoral heterogeneity of F-18 FDG uptake differentiates between gastrointestinal stromal tumors and abdominal malignant lymphomas on PET/CT. Ann Nuclear Med, 26:222–227.

31.

Nakajo

, Nakajo

, Jinguji

, Fukukura

, Nakabeppu

, Tani

, Yoshiura

. 2015. The value of intratumoral heterogeneity of (18)F-FDG uptake to differentiate between primary benign and malignant musculoskeletal tumours on PET/CT. Br J Radiol, 88:20150552.

32.

Hanaoka

, Hosono

, Tatsumi

, Ishii

, Im

, Tsuchiya

, Sakaguchi

, Matsumura

. 2015. Heterogeneity of intratumoral (111)In-ibritumomab tiuxetan and (18)F-FDG distribution in association with therapeutic response in radioimmunotherapy for B-cell non-Hodgkin's lymphoma. EJNMMI Res, 5:10.

33.

Kang

, Song

, Byun

, Oh

, Kim

, Hong

, Kwon

, Chong

, Kim

, Cho

, Park

, Kim

, Ahn

, Min

, Bom

. 2014. Intratumoral metabolic heterogeneity for prediction of disease progression after concurrent chemoradiotherapy in patients with inoperable stage III non-small-cell lung cancer. Nuclear Med Mol Imaging, 48:16–25.

34.

Tixier

, Vriens

, Cheze-Le Rest

, Hatt

, Disselhorst

, Oyen

, de Geus-Oei

, Visser

, Visvikis

. 2016. Comparison of tumor uptake heterogeneity characterization between static and parametric 18F-FDG PET images in non-small cell lung cancer. J Nucl Med, 57:1033–1039.

35.

van Velden

, Cheebsumon

, Yaqub

, Smit

, Hoekstra

, Lammertsma

, Boellaard

. 2011. Evaluation of a cumulative SUV-volume histogram method for parameterizing heterogeneous intratumoural FDG uptake in non-small cell lung cancer PET studies. Eur J Nucl Med Mol Imaging, 38:1636–1647.

36.

Chung

, Kang

, Ha

, Kim

, Park

, Song

, Cheon

. 2016. Prognostic value of preoperative intratumoral FDG uptake heterogeneity in early stage uterine cervical cancer. J Gynecol Oncol, 27:e15.

37.

van Griethuysen

JJM

, Fedorov

, Parmar

, Hosny

, Aucoin

, Narayan

, Beets-Tan

RGH

, Fillion-Robin

J-C

, Pieper

, Aerts

HJWL

. 2017. Computational radiomics system to decode the radiographic phenotype. Cancer Res, 77:e104–e107.

38.

Presotto

, Bettinardi

, De Bernardi

, Belli

, Cattaneo

, Broggi

, Fiorino

. 2018. PET textural features stability and pattern discrimination power for radiomics analysis: an “ad-hoc” phantoms study. Phys Med, 50:66–74.

39.

Zwanenburg

, Leger

, Vallières

, Lock

. 2019. Image biomarker standardisation initiative. arXiv preprint arXiv:1612.07003.

40.

Tixier

, Hatt

, Le Rest

, Le Pogam

, Corcos

, Visvikis

. 2012. Reproducibility of tumor uptake heterogeneity characterization through textural feature analysis in 18F-FDG PET. J Nucl Med, 53:693–700.

41.

Orlhac

, Soussan

, Maisonobe

J-A

, Garcia

, Vanderlinden

, Buvat

. 2014. Tumor texture analysis in 18F-FDG PET: relationships between texture parameters, histogram indices, standardized uptake values, metabolic volumes, and total lesion glycolysis. J Nucl Med, 55:414–422.

42.

De Lellis

, Lloyd

, Heitz

, Eng

. 2004 Pathology and genetics of tumour of endocrine organs. In: Kleihues

, Sobin

(eds) WHO Classification of Tumours. Vol. Lyon: IARC, pp. 49–133.

43.

Giovanella

, Ceriani

, Treglia

. 2014. Role of isotope scan, including positron emission tomography/computed tomography, in nodular goitre. Best Pract Res Clin Endocrinol Metab, 28:507–518.

Radiomics Analysis of [ 18 F]-Fluorodeoxyglucose-Avid Thyroid Incidentalomas Improves Risk Stratification and Selection for Clinical Assessment

Abstract

Background:

Methods:

Results:

Conclusions:

Introduction

Materials and Methods

Study design and patient selection

[18F]FDG-PET/CT image acquisition

[18F]FDG-PET/CT metrics

Radiomics features

Reference standard

Ethics

Statistical analyses

Results

Characteristics of patients

PET metrics and radiomics analysis

Multiparametric predictive model

Discussion

Footnotes

Author Disclosure Statement

Funding Information

References

[¹⁸F]FDG-PET/CT image acquisition

[¹⁸F]FDG-PET/CT metrics