Intratumoral and peritumoral radiomics using multi-phase contrast-enhanced CT for diagnosis of renal oncocytoma and chromophobe renal cell carcinoma: a multicenter retrospective study
PurposeTo construct diagnostic models that distinguish renal oncocytoma (RO) from chromophobe renal cell carcinoma (CRCC) using intratumoral and peritumoral radiomic features from the corticomedullary phase (CMP) and nephrographic phase (NP) of computed tomography, and compare model results with man...
Saved in:
| Main Authors: | , , , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Frontiers Media S.A.
2025-02-01
|
| Series: | Frontiers in Oncology |
| Subjects: | |
| Online Access: | https://www.frontiersin.org/articles/10.3389/fonc.2025.1501084/full |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | PurposeTo construct diagnostic models that distinguish renal oncocytoma (RO) from chromophobe renal cell carcinoma (CRCC) using intratumoral and peritumoral radiomic features from the corticomedullary phase (CMP) and nephrographic phase (NP) of computed tomography, and compare model results with manual and radiological results.MethodsThe RO and CRCC cases from five centers were split into a training set (70%) and a validation set (30%). CMP and NP intratumoral and peritumoral (1–3 mm) radiomic features were extracted. Segmentation was performed by radiologists and software. Features with high intraclass correlation coefficients (ICC>0.75) were selected through univariate analysis, followed by the LASSO method to determine the final features for the SVM model. All images were assessed by two radiologists, and radiological reports were also examined. The diagnostic performances of the different methods were compared using several statistical methods.ResultsThe training set had 65 cases (29 RO, 36 CRCC) and the validation set had 27 cases (12 RO, 15 CRCC). All the training models had excellent performance (area under the curve [AUC]: 0.828–0.942); the AUC values of the validation models ranged from 0.900 (Model 4) to 0.600 (Model 2). CMP models (AUC: 0.811–0.900) generally outperformed NP and fusion models (AUC: 0.728–0.756). SVM models (sensitivity: 62.50–88.89%; specificity: 63.16–77.78%; accuracy: 62.96–81.48%) outperformed manual diagnosis (sensitivity: 46.74–70.59%; specificity: 41.67–46.34%; accuracy: 52.27–59.78%). The clinical reports alone had no diagnostic value.ConclusionCMP intratumoral and peritumoral radiomics models reliably distinguished RO from CRCC. |
|---|---|
| ISSN: | 2234-943X |