Radiomics feature robustness of photon-counting CT and DECT in chest tumor phantom: A dual-center, multi-device study.

Abstract

e20011 Background: Robustness of radiomics features across CT platforms is essential for clinical translation of quantitative imaging biomarkers. This study evaluated radiomics feature robustness derived from photon-counting detector CT (PCD-CT) and multiple dual-energy CT (DECT) systems using standardized chest tumor phantoms. Methods: Twelve chest tumor phantoms with different sizes and texture patterns were constructed using a CRIS chest phantom. All phantoms were scanned at two centers on four energy CT platforms: PCD-CT, dual-layer DECT, dual-source DECT, and rapid kV-switching DECT. Images were reconstructed as conventional images, virtual monoenergetic images at 40, 70, and 100 keV, and virtual non-contrast images. Tumor-specific regions of interest were delineated, and radiomics features were extracted after standardized preprocessing. Inter-scanner robustness was assessed using concordance correlation coefficient, coefficient of variation, and quartile coefficient of dispersion. Results: Radiomics feature robustness varied substantially by reconstruction type. In conventional images, the highest inter-scanner agreement was observed between dual-source and rapid kV-switching DECT, whereas most other device pairs showed limited robustness. Among texture phantoms, acceptable agreement was additionally observed between PCD-CT and dual-layer DECT. In virtual non-contrast images, robustness between PCD-CT and DECT platforms was consistently poor, while agreement among DECT systems remained acceptable. Low-energy virtual monoenergetic imaging at 40 keV significantly improved robustness between PCD-CT and DECT platforms, whereas robustness decreased at higher energy levels. Similar trends were observed across different radiation dose settings. Conclusions: Radiomics feature robustness across modern energy CT platforms is highly dependent on reconstruction strategy. Low-energy virtual monoenergetic imaging improves inter-scanner robustness, including for PCD-CT, whereas virtual non-contrast reconstruction compromises harmonization between PCD-CT and DECT systems. These findings support reconstruction selection and radiomics standardization in multicenter thoracic oncology studies.