A Computed Tomography-Based Score to Predict Survival in Patients With Adrenocortical Carcinoma: A Proof-of-Concept Study
Canadian Association of Radiologists Journal 2025, Vol. 76(4) 683-691 @ The Author(s) 2025 Article reuse guidelines: sagepub.com/journals-permissions DOI: 10.1177/08465371251335170 journals.sagepub.com/home/caj
Sage
Maxime Barat1,2(D, Mohamed Eltaher3, Ahmed W. Moawad4, Philippe Soyer1,2 D, David Fuentes3, Marianne Golse1,2, Anne Jouinot2,5, Ayahallah A. Ahmed3, Mostafa A. Shehata3, Guillaume Assié2,5, Mohab M. Elmohr6, Magalie Haissaguerre7, Mouhammed A. Habra3, Christine Hoeffel8, Khaled M. Elsayes3, Jérome Bertherat2,5, and Anthony Dohan1,2
Abstract
Purpose: Adrenocortical carcinoma (ACC) is a rare condition with a poor and hardly predictable prognosis. This study aims to build and evaluate a preoperative computed tomography (CT)-based score (CT score) using features previously reported as biomarkers in ACC to predict overall survival (OS) in patients with ACC. Methods: A CT score based on preoperative CT examinations combining shape elongation, maximum tumour diameter, and the European Network for the Study of Adrenal Tumors (ENSAT) stage was built using a logistic regression model to predict OS duration in a development cohort of 89 patients with ACC. An optimal cut-off of the CT score was defined and the Kaplan-Meier method was used to assess OS. The CT score was then tested in an external validation cohort of 54 patients wit ACC. The C-index of the CT score for predicting OS was compared to that of ENSAT stage alone. Results: The CT score helped discriminate between patients with poor prognosis and patients with good prognosis in both the validation cohort (54 patients; mean OS, 69.4months; 95% confidence interval [CI]: 57.4-81.4months vs mean OS, 75.6 months; 95% CI: 62.9-88.4 months, respectively; P =. 022). In the validation cohort the C-index of the CT score was significantly better than that of the ENSAT stage alone (0.62 vs 0.35; P =. 002). Conclusion: A CT score combining morphological criteria, radiomics, and ENSAT stage on preoperative CT examinations allows a better prognostic stratification of patients with ACC compared to ENSAT stage alone.
Résumé
Objectif: Le carcinome corticosurrénalien (CCS) ou corticosurrénalome est une affection rare, caractérisée par un pronostic défavorable et difficilement prévisible. La présente étude visait à concevoir et à évaluer un score basé sur la tomodensitométrie (TDM) préopératoire (score TDM), en utilisant des caractéristiques auparavant rapportées comme biomarqueurs du CCS, afin de prédire la survie globale (SG) chez les patients atteints de CCS. Méthodes : Un score TDM a été développé à partir d’examens de TDM préopératoires, en combinant l’allongement, le diamètre maximal et le stade ENSAT (European Network for the Study of Adrenal Tumors) de la tumeur. Ce score a été construit au moyen d’un modèle de régression logistique afin de prédire la durée de la survie globale (SG) dans une cohorte de développement de 89 patients avec un CCS. Un seuil optimal du score TDM a été déterminé, et la méthode de Kaplan-Meier a été utilisée pour évaluer la SG. Le score TDM a ensuite été testé dans une cohorte de validation externe de 54 patients avec un CCS. L’indice de concordance du score TDM pour la prédiction de la SG a été comparé à celui du stade ENSAT seul. Résultats : Dans la cohorte de validation, le score TDM a permis de distinguer les patients ayant un mauvais pronostic (SG moyenne: 69,4 mois; intervalle de confiance [IC] à 95 %: 57,4-81,4 mois) de ceux ayant un bon pronostic (SG moyenne: 75,6 mois; IC à 95 %: 62,9- 88,4 mois) (54 patients; P=0,022). En outre, l’indice de concordance du score TDM (0,62) était significativement supérieur à celui du stade ENSAT seul (0,35) (P=0,002). Conclusion: Un score TDM combinant des critères morphologiques, des données issues de la radiomique et le stade ENSAT à partir d’examens préopératoires permet une stratification pronostique plus précise que le stade ENSAT seul chez les patients atteints de CCS.
Keywords
Multidetector computed tomography, Prognosis, Adrenocortical carcinoma, Image processing, Computer-assisted
Introduction
Adrenocortical carcinoma (ACC) is a rare condition with an estimated incidence of 0.5 to 2 ACCs per million of inhabit- ants per year, and it accounts for 0.04% to 0.2% of all cancer deaths in the United States.1,2 Patients with ACC, generally have a poor prognosis, with a 5-year overall survival (OS) rate of only about 40%. However, OS varies considerably among patients.3 In patients with ACC, prognostic factors for OS include clinical, histopathological, and molecular fea- tures; these factors could help determine appropriate manage- ment based on patient prognosis,4 but histopathological and molecular information are obtained at the penalty of invasive tissue sampling, which is not free of morbidity5 and carries a risk of tumour dissemination.6 Therefore, adrenal biopsy is not recommended in the setting of ACC.7 Among the histo- pathological factors that have demonstrated prognostic capa- bilities is the Ki-67 index.8 A Ki-67 index >10% is regarded as a marker of high risk of recurrence, justifying the use of adjuvant chemotherapy in these patients.8
The preoperative prognosis of patients with ACC is usu- ally assessed using the European Network for the Study of Adrenal Tumors (ENSAT) staging system, which is based on magnetic resonance imaging (MRI) of the adrenal gland and computed tomography (CT) examination of the chest abdomen and pelvis (Table 1).8 Conventional imaging alone is not sufficient to diagnose, classify, or estimate the prog- nosis of patients with ACC9; therefore, the ENSAT classifi- cation takes into account tumour size, infiltration of surrounding adipose tissue, invasion of adjacent organs, positive lymph nodes, and distant metastases.8 It provides a stratification strongly associated with cancer-specific mor- tality; the 5-year OS is >60% for patients with ENSAT stage I or II, <50% for patients with stage III, and 20% for patients with stage IV ACC.10,11 However, the high hetero- geneity of OS especially in patients with ENSAT stage I to III may require improvements.
Transcriptomics, which is an unsupervised high-through- put analysis of tumour genome expression profile,12 can also be used to predict outcomes in patients with ACC. This analy- sis is performed on histopathological tissue samples, and the results correlate with patient disease-free survival (DFS) and OS.13 In ACC, transcriptomics identifies two genomic pro- files that correspond to two groups of patients with different outcomes.13 The C1A profile is associated with a poor
| ENSAT stage | Definition |
|---|---|
| I | T1, N0, M0 |
| II | T2, N0, M0 |
| III | T1-T2, N1, M0 |
| T3-T4, N0-N1, M0 | |
| IV | T1-T4, N0-N1, M1 |
Note. T stage. T1 corresponds to tumour ≤5cm; T2 corresponds to tumour >5 cm; T3 corresponds to infiltration into surrounding tissue; T4 corresponds to tumour invasion into adjacent organs or venous tumour thrombus in the inferior vena cava or renal vein. N stage. N0 corresponds to no positive lymph nodes; N1 corresponds to positive lymph nodes. M stage. M0 corresponds to no distant metastases; M1 corresponds to the presence of distant metastases.
outcome, with a 5-year OS rate approaching 0%, whereas the C1B profile is associated with a 5-year OS rate >90%.13 However, this analysis requires invasive tissue sampling with a potential associated morbidity.
To improve the preoperative and non-invasive evaluation of patients with ACC, the radiomic approach has been consid- ered.8 Radiomics is the analysis of mathematically derived textural features on radiological images.14 Radiomics has demonstrated capabilities in the field of adrenal lesion char- acterization15-17 and can also be used to predict OS and DFS in other cancers.18,19 One study specifically evaluated the capabilities of radiomics for estimating Ki-67 index in ACC.20 This study found that 2 shape-based features (ie, shape elon- gation [SE] and shape flatness [SF]) were predictive of Ki-67 index >10%.20 SE >0.668 was predictive of a Ki-67 index >10% with sensitivity, specificity, and positive likelihood and negative likelihood ratios of 75%, 69.2%, 2.4, and 0.36, respectively. SF >0.4966 was a predictor of a Ki-67 index
10% with sensitivity, specificity, and positive likelihood and negative likelihood ratios of 80%, 69.2%, 2.5, and 0.28, respectively. However, the study did not evaluate to what extent these features could be used as predictors of patient survival.20
The aim of this study was to evaluate the performance of a preoperative computed tomography (CT)-based radiomic prognostic score using features previously reported as predic- tors of high Ki-67 index (ie, Ki-67 index >10%) in ACC to predict OS in patients with this condition.
’ Department of Radiology, Hôpital Cochin, AP-HP, Paris, France
2 Université Paris Cité, Faculté de Médecine, Paris, France
3 The University of Texas MD Anderson Cancer Center, Houston, TX, USA
4 Mercy Catholic Medical Center, Darby, PA, USA
5 Department of Endocrinology, Hôpital Cochin, AP-HP, Paris, France
6 Baylor College of Medicine, Houston, TX, USA
7 Department of Endocrinology, Universitaire Hôpital de Bordeaux, Pessac, France
8 Reims Medical School, Department of Radiology, Hôpital Robert Debré, CHU Reims, Université Champagne-Ardennes, Reims, Grand Est, France
Corresponding Author:
Email: maxime.barat@aphp.fr
708 patients with adrenalectomy from 2007 to 2022
Excluded patients (n= 599; 84.6%) Other histopathological results than adrenocortical carcinoma
Excluded patients (n = 19; 26.8%) Incomplete CT data (no venous phase CT images available)
Excluded patient (n= 1; 0.1%) Coexisting benign adenoma and adrenocortical carcinoma in the same gland
89 eligible patients with adrenocortical carcinoma and complete data
Materials and Methods
Patients
This retrospective study was approved by the institutional review board of Centre 1 (Nº: AAA-2020-08048), and the requirement for written informed consent was waived.
The database of the department of pathology of our institu- tion (Centre 1) was queried to identify all patients who under- went adrenalectomy from January 2007 to December 2022. This initial search retrieved 708 patient records. Patients were included in the study if they were older than 18 years, had a histopathologically proven diagnosis of ACC, and had a pre- operative CT examination performed less than 3 months before surgery that was available for review. Patients with adrenal tumours other than ACC (n=599), patients with col- lision tumours (n=1), and those without a preoperative CT examination (n=19) were excluded. Figure 1 shows the study flow chart of patients who were eligible for this study. Patients’ clinical and histopathological data were recorded, including age, sex, tumour secretion, largest tumour diameter, tumour side, Ki-67 index, Weiss pathological score, and ENSAT stage. The characteristics of patients in the develop- ment cohort are summarized in Table 2.
CT Data Acquisition
CT acquisition parameters are summarized in Table 3. CT examinations covered the thorax, abdomen, and pelvis and were obtained before and after intravenous administration of iodinated contrast material during the portal venous (60- 70 seconds) and delayed (10 minutes) phases.
Data Analysis
One radiologist blinded to the clinical outcomes (M. B.) with 10 years of experience in abdominal imaging) reviewed all CT examinations after anonymization on a viewing station of the picture-archiving and communication system of our insti- tution (DirectView®, 11.4.0.1253 sp1 version, Carestream Health, Rochester, NY, USA). The morphological character- istics of the adrenal lesions, including maximum tumour diameter measured in the axial plane, vascular involvement, and all ENSAT stage criteria, were determined by this radi- ologist based on review of the CT examinations.
Three-dimensional segmentation of CT images of the entire adrenal lesion was performed using the ITKsnap v1.0.0rc2 module of the 3D-Slicer® software (https://www. slicer.org) by the same radiologist with the instruction to delineate a volume that included the entire adrenal lesion tis- sue while excluding the healthy gland and periadrenal tissue. This approach was intended to minimize contamination from the peritumoral environment and non-tumoral tissues caused by partial volume effects, and to ensure accurate extraction of radiomic features (Figure 2). Shape radiomic features were extracted from CT data obtained during the portal venous phase of enhancement using the PyRadiomic module (http:// www.radiomics.io/pyradiomics.html) of the same software after normalization of the voxel size at 1 × 1 × 1 mm3.21 The PyRadiomics is an open-source Python package that extracts a wide range of radiomic features from any medical images. These features include first-order statistics (ie, histogram), shape-based features (eg, volume, sphericity), and texture features derived from different matrix after matricial transfor- mation. PyRadiomics is widely used in medical research for
| Variable | Development cohort (n=89) | Validation cohort (n=54) | P value* |
|---|---|---|---|
| Male | 26/89 (29%) | 22/54 (41%) | .20 |
| Right-sided ACC | 47/89 (53%) | 21/54 (39%) | .11 |
| Secretion | 56/89 (63%) | 26/54 (48%) | .08 |
| ENSAT stage | .08 | ||
| I | 10/89 (11%) | 4/54 (7.4%) | |
| II | 45/89 (51%) | 19/54 (35%) | |
| III | 19/89 (21%) | 22/54 (41%) | |
| IV | 15/89 (17%) | 9/54 (17%) | |
| Ki67 rate (%) | 22 ±19 [0-95] | 25 ± 17 [2-79] | .07 |
| Weiss score | 6.53 ± 1.96 [3.00-9.00] | 5.83 ± 1.57 [3.00-9.00] | .06 |
| Follow up (mo) | 51 ±46 [1-176] | 69 ±42 [7-206] | <. 01 |
| Death | 28/89 (31%) | 29/54 (54%) | <. 01 |
| Shape elongation | 0.80 ±0.11 [0.54-0.97] | 0.66 ±0.25 [0.01-0.96] | <. 01 |
| Largest ACC diameter (mm) | 82 ±45 [10-219] | 116 ±80 [18-368] | .03 |
Note. Qualitative variables are expressed as proportions followed by percentages into parentheses. Quantitative variables are expressed as
means + standard deviations followed by ranges into brackets. ACC =adrenocortical carcinoma; ENSAT = European Network for the Study of Adrenal Tumors.
*Comparison between development and validation cohorts. Bold indicates significant P value.
| Variable | Training cohort | Validation | ||
|---|---|---|---|---|
| Equipment | Somatom Sensation® 64 (Siemens Healthineers) | Somatom Definition® Flash (Siemens Healthineers) | Revolution HD (General Electrics) | CT Light-Speed (General Electrics) |
| Number of detectors | 64 | 64 | 128 | 64 |
| Slice thickness acquisition (mm) | 0.6-1.25 | 0.6-1.25 | ||
| Slice thickness reconstruction (mm) | 1-3 | 2.5 | ||
| Rotation time (s) | 0.5-0.7 | 0.8 | ||
| Peak tube potential (kVp) | 110-120 | |||
| Field-of-view (mm) | 279-350 | |||
| Contrast agent | lomeron® 350ª or Xenetix® 350b | Omnipaque ™ 300℃ | ||
| Injection rate (mL/s) | 2.5-4 | 3-5 | ||
a Iomeprol, Iomeron 350® (Bracco Imaging, Milan, Italy).
blobitridol, Xenetix 350® (Guerbet, Aulnay-sous-bois, France).
clohexol, Omnipaque 300™ (General Electric, Chicago, III, USA).
tasks such as tumour characterization, disease classification, and treatment prediction, offering quantitative insights that aid clinical decision-making. For this study, only 16 shape features were extracted.21 These features were analyzed with- out any post-extraction processing.
Validation Cohort
For the validation cohort, 54 patients from another centre (Centre 2) were retrospectively identified from a previously published study.20 Similar inclusion and exclusion criteria were applied. Patients medical records were retrospectively analyzed, and age, sex, Ki-67 index, ENSAT stage, shape elongation, occurrence of death, and duration of follow-up were recorded.
In this cohort, preoperative CT images were obtained in the venous phase (60-80 seconds after intravenous adminis- tration of iodinated contrast material). Some of the CT exami- nations included in the present study were performed at outside facilities. CT acquisition parameters are summarized in Table 3.20 Segmentation was performed using the same method as described above.
Statistical Analysis
Statistical analysis was performed using R software (version 4.1.0, R-foundation, http://www.r-project.org/). Quantitative (continuous) variables were reported as means ± standard deviations (SD) and ranges. Qualitative (binary) variables were reported as raw numbers, proportions, and percentages.
1
5: 15:5.9580m
A
B
B:
ć
A: 108.8912mm * Y 90
R: 59.4085mm
Continuous variables were compared using Student t-test, and categorical variables using ×2 test or Fisher exact test, as appropriate.22
Thirty-five ACC images were randomly selected from the development cohort to assess interrater agreement. A radiolo- gist (M. G.) with 10 years of experience in abdominal imag- ing performed new segmentations for the selected ACC images while blinded to the segmentation results of the previ- ous radiologist. The Dice similarity coefficient (DSC) was calculated to evaluate interobserver reproducibility of the segmentation methods for each ACC.23 Intraclass correlation coefficients (ICCs) were calculated for each radiomic feature using a two-way mixed effects model.24 The ICC is a value between 0 and 1, with values below .5 indicating poor reli- ability, between .5 and .75 indicating moderate reliability, between .75 and .9 indicating good reliability, and any value above .9 indicating excellent reliability.25 Features with an ICC <. 9 were considered non-reproducible.26,27
Among previously published features (SE and SF) that correlated with a Ki-67 index >10% in the study by Ahmed et al, only SE with an ICC >.9 were selected.2º A receiver operating characteristic (ROC) curve was built, and area under the curve (AUROC) for the prediction of Ki-67 index >10% was calculated.
Of the features selected by Ahmed et al, only shape elon- gation was retained and added to other clinico-radiological features. A multivariable cox proportional hazards model was
used to select independent predicting variables using the 24 months OS as explained variable.
Then, a multivariable model was then developed using a logistic regression method based on the development cohort. Features selected were maximum tumour diameter, ENSAT stage, and shape elongation.20 The model was tested for its ability to predict patient OS using a ROC curve. The optimal cut-off value, defined as the value of the features that yielded the best accuracy for diagnosing of a high Ki-67 index as identified by the ROC curve analysis, was used to divide the cohort into patients with a good prognosis (ie, OS >24 months) and those with a poor prognosis (ie, OS <24months). Patients’ OS was estimated by using the Kaplan-Meier method. The log-rank test was used to compare OS between patients with good prognosis and those with poor prognosis. The CT-based radiomic model was then tested in the valida- tion cohort using the same method.
The performance of the radiomic score in predicting OS was assessed using Harrel’s C-index and compared to the ENSAT stage alone in the validation cohort using the Student t-test for the comparison of C-indexes. All tests were two- tailed, and significance was set at P <. 05.
No missing data were observed for the features included in the CT-based radiomic model in either the development or validation cohorts.
A nomogram for predicting mortality using the CT-based radiomic model was built based on the final multivariable cox
1.0
0.8
0.562 (0.611, 0.849)
0.6
Sensitivity
AUC: 0.779
0.4
0.2
0.0
1.0
0.8
0.6
0.4
0.2
0.0
Specificity
proportional hazards model using the respective coefficients of each independent predictor.
Results
Eighty-nine patients were included in the development cohort and 54 patients in the validation cohort. Patient characteris- tics are summarized in Table 2.
Segmentation and extraction of radiomic features was possible for all ACC images. The reproducibility of segmen- tation was considered almost perfect for 34/35 (97%) ACCs with a mean DSC of 0.92 ± 0.03 (SD; range: 0.72-0.97). SF had an ICC <. 9 (ICC =. 85) and was excluded from the anal- ysis. Therefore, the only radiomic feature used in the analy- sis was SE.
In the development cohort, the Ki-67 index was available for 77 of 89 patients (87%). The AUROC for predicting Ki-67 index >10% was 0.64 (95% CI: 0.54-0.76).
The CT score developed in the development cohort included SE, maximum tumour diameter, and ENSAT stage as assessed using pre-treatment CT examination. The AUROC for predicting patients’ OS was 0.779, and the best cut-off value for dividing the development cohort into patients with OS >24 months and those with OS <24 months was 0.562 (Figure 3). The Kaplan Meier analysis in the development cohort yielded a mean OS of 27.1 months (95% CI: 13.7- 40.1 months) in patients with poor prognosis and 63.3 months (95% CI: 50.1-76.5 months) in those with good prognosis (P <. 001; Figure 4).
8
8
Overall survival (%)
0
4
&
0
Logrank P = 1,1E-08
0
12
24
36
48
60
72
84
96
No. At Risk
Time from surgery (months)
Radscore $ 0.562
Radscore > 0.562
30
18
8
4
4
4
59
48
45
4
36
30
4
25
22
19
17
0
8
Overall survival (%)
8
$
₹
0
Logrank P = 2.2E-02
0
12
24
36
Time from surgery (months)
48
60
72
84
96
No. At Risk Radscore $ 0.562
Radscore > 0,562
28
23
21
16
26
14
26
25
21
8
20
12
19
11
13
12
6
10
In the validation cohort, the CT score helped discrimi- nate between patients with poor prognosis (mean OS, 69.4 months; 95% CI: 57.4-81.4 months) and those with good prognosis (mean OS, 75.6 months; 95% CI: 62.9- 88.4 months; P =. 022; Figure 5). The C-index of the CT score was significantly better than that of the ENSAT stage alone (0.62 vs 0.35, respectively; P =. 002) in the valida- tion cohort (Figure 6).
The nomogram for predicting mortality is summarized in Figure 7. Adrenal tumour size, ENSAT stage, and shape elon- gation were used to build the nomogram. For each patient, points are assigned based on each of these 3 predictors using the top section of the nomogram and are summed. The total points so generated are then related to predicted 3-, 5-, and 10-year mortality probability in the bottom section.
0
Overall survival (%)
0
$
&
o
Logrank P = 1.7E-01
0
12
24
36
48
Time from surgery (months)
60
72
84
96
No. At Risk ENSAT 1-2 ENSAT 3-4
23
23
22
18
17
17
16
10
31
26
24
19
15
14
8
12
Discussion
This retrospective, two-centre study reveals that a CT score including SE, maximum tumour diameter, and ENSAT stage helps distinguish between patients with ACC who have a good prognosis and those with poor prognosis in an external validation cohort.
As in the study by Ahmed et al, tumour segmentation, radiomic feature extraction, and analysis were performed using CT data obtained from portal venous phase images only.20 For most patients with suspected adrenal disease, CT is the first-line imaging modality for adrenal lesion character- ization, whereas MRI is primarily used for specific indica- tions.28,29 In addition, the portal venous phase is the most frequently performed acquisition phase for oncologic staging, and results in homogeneous contrast agent distribution, which may potentially improve features extraction reproducibility.
SE is a radiomic feature consisting of the ratio of the minor axis length to the major axis length of the tumour.21 SE on pre- operative imaging was found to correlate with a Ki-67 index cut-off value of 10%.20 The combination of SE with existing prognostic factors on preoperative imaging (ie, ENSAT stage and maximum diameter) may improve patient care by providing accurate risk stratification and enabling aggressive treatment, if needed, immediately after diagnosis. Moreover, using a manual segmentation method, SE extraction was highly reproducible, with an ICC >.9.24 A manual measurement of the largest and smallest diameters could enable the calculation of “shape elon- gation” without the need for labour-intensive segmentation. However, the precision of these measurements would greatly affect the results, and likely have a substantial negative impact on the stability of the parameter.
Shape-based radiomic features have demonstrated utility for tumour characterization in oncologic imaging and for prognostic stratification of patients.30 Moreover, a high
interobserver reproducibility between radiologists has been reported.30 A shape-based classification using 3 features (roughness, convexity, and sphericity) proved to be able to discriminate between lung granuloma and lung carcinoma with an AUROC of 0.72, which was equivalent to that of expert radiologists.3º Similarly, Alvarez-Jimenez et al showed that first-order and shape-based radiomic features measured on T2-weighted MRI images of the rectal wall and peritu- moral environment of rectal cancer after neoadjuvant radio- chemotherapy were effective for the restaging of rectal tumours before surgery.31 In this study, radiomic features were associated with tumour grade after radiochemotherapy with an accuracy of 69% for the detection of ypT0-2 versus ypT3-4 tumours using data from 52 patients; in an external validation cohort of 42 patients, the detection accuracy was 62%.31 The study by Ahmed et al, also identified SF as a pre- dictor of Ki67 tumour rate; however, no stability test was con- ducted.20 Based on the ICC results in our study, SF was considered unstable and therefore excluded from the analysis in the present study.27
Regarding the application of radiomics for AL characteriza- tion, most studies included few patients and no validation cohort. For example, Feliciani et al included 48 patients to differentiate benign from malignant AL with spontaneous attenuation greater than 10 HU on preoperative CT and found an AUROC of 0.93.32 Umanodan et al and Kong et al included 52 and 305 patients, respectively, to characterize pheochromocytoma using radiomics applied to MRI, but did not consider biological tests, which is a relevant point for the diagnosis of pheochromocytoma.33,34 To our knowledge, no studies have aimed to evaluate the prognosis of AL using a radiomic model.
Currently, adjuvant treatments such as mitotane are indi- cated in patients with ACC at high risk of recurrence based on surgical and histopathological findings including complete- ness of the resection, tumour grade, and Ki-67 index.8 When indicated, adjuvant treatment should be started as soon as possible.8 The significantly higher C-index of our model compared to the ENSAT stage alone in the validation cohort suggests improved discriminating capabilities of our CT score over pre-existing predictors. This improvement allows decision makers to more confidently stratify patients into risk categories and effectively prioritize adjuvant treatments.
Our study has some limitations. First, despite a two-centre design, the number of patients remains limited, mostly because of the rarity of ACC, which makes the performance of large studies difficult. However, our study includes a rela- tively large number of patients compared to previous pub- lished studies, as well as an external validation cohort.20 Second, the retrospective design of the study may introduce selection bias, but patients with this rare disease are mostly treated in reference centres that belong to the COMETE- cancer network, so their treatment is largely homogeneous.35 Similarly, the acquisition parameters of the 2 centres were different, which might have influenced the results of the extracted radiomic features; however, the use of voxel size normalization for radiomic analysis should mitigate this risk.
Points
0
1
2
3
4
5
6
7
8
9
10
Size (mm)
0
20
60
100
140
180
220
ENSAT>3
1
0
Shape Elongation
0 0.1
0.3
0.5
0.7
0.9
1
Total Points
0
2
4
6
8
10
12
14
16
18
3-year survival
0.85
0.8
0.75
0.7
0.6
0.5
5-year survival
0.8
0.75
0.7
0.6
0.5
10-year survival
0.75
0.7
0.6
0.5
Finally, the retrospective design also limits the OS analysis because a subset of patients was lost to follow-up.
In conclusion, we have developed a CT score based on preoperative CT images that can help discriminate between ACC patients with good and poor prognosis in an external validation cohort. These results may improve the periopera- tive management and risk stratification of patients with ACC but need to be confirmed by further prospective studies.
Abbreviations
ACC Adrenocortical carcinoma
AUC Area under receiver operating characteristic curve
CI Confidence interval
CT computed tomography
DFS Disease-free survival
DSC Dice similarity coefficient
ENSAT European Network for the Study of Adrenal Tumors ICC Intraclass correlation coefficient
OS Overall survival
SD Standard deviation
SE Shape elongation
SF Shape flatness
Acknowledgments
Maxime Barat had a research grant in relation with this manuscript from the Société Française de Radiologie and the Servier Institute.
Data Availability Statement
Data are available on demand.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors received no financial support for the research, author- ship, and/or publication of this article.
Human Rights
The authors declare that the work described has been performed in accordance with the Declaration of Helsinki of the World Medical Association revised in 2013 for experiments involving humans.
Ethics Approval
This retrospective study was approved by the institutional review board of Centre 1 (Nº: AAA-2020-08048).
Informed Consent
The requirement for written informed consent was waived by Institutional Review Board.
ORCID iDs
Maxime Barat ID https://orcid.org/0000-0002-0360-0875
Philippe Soyer ID https://orcid.org/0000-0002-5055-1682
References
1. Souteiro P, Donato S, Costa C, et al. Diagnosis, treatment, and survival analysis of adrenocortical carcinomas: a multicentric study. Hormones. 2020;19(2):197-203.
2. Abiven G, Coste J, Groussin L, et al. Clinical and biological fea- tures in the prognosis of adrenocortical cancer: poor outcome of cortisol-secreting tumors in a series of 202 consecutive patients. J Clin Endocrinol Metab. 2006;91(7):2650-2655.
3. Grubbs E, Lee JE. Limited prognostic value of the 2004 International Union Against Cancer staging classification for adrenocortical carcinoma: proposal for a revised TNM classifi- cation. Cancer. 2009;115(24):5847.
4. Assie G, Jouinot A, Fassnacht M, et al. Value of molecular classification for prognostic assessment of adrenocortical carci- noma. JAMA Oncol. 2019;5(10):1440-1447.
5. Bancos I, Tamhane S, Shah M, et al. Diagnosis of endo- crine disease-the diagnostic performance of adrenal biopsy: a systematic review and meta-analysis. Eur J Endocrinol. 2016;175(2):R65-R80.
6. Williams AR, Hammer GD, Else T. Transcutaneous biopsy of adrenocortical carcinoma is rarely helpful in diagnosis, potentially harmful, but does not affect patient outcome. Eur J Endocrinol. 2014;170(6):829-835.
7. Fassnacht M, Arlt W, Bancos I, et al. Management of adre- nal incidentalomas: European society of endocrinology clini- cal practice guideline in collaboration with the European Network for the Study of Adrenal Tumors. Eur J Endocrinol. 2016;175(2):G1-G34.
8. Fassnacht M, Dekkers OM, Else T, et al. European Society of Endocrinology Clinical Practice Guidelines on the management of adrenocortical carcinoma in adults, in collaboration with the European Network for the Study of Adrenal Tumors. Eur J Endocrinol. 2018;179(4):G1-G46.
9. Barat M, Gaillard M, Cottereau AS, et al. Artificial intelligence in adrenal imaging: a critical review of current applications. Diagn Interv Imaging. 2023;104(1):37-42.
10. Lughezzani G, Sun M, Perrotte P, et al. The European Network for the Study of Adrenal Tumors staging system is prog- nostically superior to the international union against cancer- staging system: a North American validation. Eur J Cancer. 2010;46(4):713-719.
11. Kedra A, Dohan A, Gaujoux S, et al. Preoperative detection of liver involvement by right-sided adrenocortical carcinoma using CT and MRI. Cancers. 2021;13(7):1603.
12. Barat M, Crombé A, Boeken T, et al. Imaging in France: 2024 update. Can Assoc Radiol J. 2025;76(2):221-231.
13. Assie G, Letouze E, Fassnacht M, et al. Integrated genomic characterization of adrenocortical carcinoma. Nat Genet. 2014;46(6):607-612.
14. Javed AA, Zhu Z, Kinny-Köster B, et al. Accurate non-invasive grading of nonfunctional pancreatic neuroendocrine tumors with a CT derived radiomics signature. Diagn Interv Imaging. 2024;105(1):33-39.
15. Yi X, Guan X, Zhang Y, et al. Radiomics improves effi- ciency for differentiating subclinical pheochromocytoma from lipid-poor adenoma: a predictive, preventive and personal- ized medical approach in adrenal incidentalomas. EPMA J. 2018;9(4):421-429.
16. Elmohr MM, Fuentes D, Habra MA, et al. Machine learning- based texture analysis for differentiation of large adrenal corti- cal tumours on CT. Clin Radiol. 2019;74(10):818.e1-818.e7.
17. Yi X, Guan X, Chen C, et al. Adrenal incidentaloma: machine learning-based quantitative texture analysis of unenhanced CT can effectively differentiate sPHEO from lipid-poor adrenal adenoma. J Cancer. 2018;9(19):3577-3582.
18. Masson-Grehaigne C, Lafon M, Palussière J, et al. Single- and multi-site radiomics may improve overall survival prediction for patients with metastatic lung adenocarcinoma. Diagn Interv Imaging. 2024;105(11):439-452.
19. Liu P, Tan XZ, Zhang T, et al. Prediction of microvascular invasion in solitary hepatocellular carcinoma ≤ 5 cm based
on computed tomography radiomics. World J Gastroenterol. 2021;27(17):2015-2024.
20. Ahmed AA, Elmohr MM, Fuentes D, et al. Radiomic mapping model for prediction of Ki-67 expression in adrenocortical car- cinoma. Clin Radiol. 2020;75:479.e17-479.e22.
21. van Griethuysen JJM, Fedorov A, Parmar C, et al. Computational radiomics system to decode the radiographic phenotype. Cancer Res. 2017;77(21):e104-e107.
22. O’Shaughnessy E, Detrinidad E, Soyer P, Lecler A. An intro- ductory guide to statistics for the radiologist. Diagn Interv Imaging. 2025;106(2):49-52.
23. Molière S, Hamzaoui D, Granger B, et al. Reference stan- dard for the evaluation of automatic segmentation algorithms: quantification of inter observer variability of manual delin- eation of prostate contour on MRI. Diagn Interv Imaging. 2024;105(2):65-73.
24. Benchoufi M, Matzner-Lober E, Molinari N, Jannot AS, Soyer P. Interobserver agreement issues in radiology. Diagn Interv Imaging. 2020;101(10):639-641.
25. Bobak CA, Barr PJ, O’Malley AJ. Estimation of an inter-rater intra-class correlation coefficient that overcomes common assumption violations in the assessment of health measurement scales. BMC Med Res Methodol. 2018;18(1):93.
26. Aerts HJ, Velazquez ER, Leijenaar RT, et al. Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach. Nat Commun. 2014;5:4006.
27. Lambin P, Leijenaar RTH, Deist TM, et al. Radiomics: the bridge between medical imaging and personalized medicine. Nat Rev Clin Oncol. 2017;14(12):749-762.
28. Haider MA, Ghai S, Jhaveri K, Lockwood G. Chemical shift MR imaging of hyperattenuating (>10 HU) adrenal masses: does it still have a role? Radiology. 2004;231(3):711-716.
29. Seo JM, Park BK, Park SY, Kim CK. Characterization of lipid- poor adrenal adenoma: chemical-shift MRI and washout CT. AJR Am J Roentgenol. 2014;202(5):1043-1050.
30. Barat M, Greffier J, Si-Mohamed S, et al. CT imaging of the pancreas: a review of current developments and applications. Can Assoc Radiol J. Published online February 22, 2025. doi:10.1177/08465371251319965
31. Alvarez-Jimenez C, Antunes JT, Talasila N, et al. Radiomic texture and shape descriptors of the rectal environment on post- chemoradiation T2-weighted MRI are associated with patho- logic tumor stage regression in rectal cancers: a retrospective, multi-institution study. Cancers. 2020;12(8):2027.
32. Feliciani G, Serra F, Menghi E, et al. Radiomics in the char- acterization of lipid-poor adrenal adenomas at unenhanced CT: time to look beyond usual density metrics. Eur Radiol. 2024;34(1):422-432.
33. Umanodan T, Fukukura Y, Kumagae Y, et al. ADC histogram analysis for adrenal tumor histogram analysis of apparent diffu- sion coefficient in differentiating adrenal adenoma from pheo- chromocytoma. J Magn Reson Imaging. 2017;45(4):1195-1203.
34. Kong J, Zheng J, Wu J, et al. Development of a radiomics model to diagnose pheochromocytoma preoperatively: a multicenter study with prospective validation. J Transl Med. 2022;20(1):31.
35. Hescot S, Debien V, Hadoux J, et al. Outcome of adrenocortical carcinoma patients included in early phase clinical trials: results from the French network ENDOCAN-COMETE. Eur J Cancer. 2023;189:112917.