2977 - Radiomics Feature Robustness of Photon-Counting CT and DECT in Chest Tumor Phantom: A Dual-Center, Multi-Device Study

Purpose/Objective(s): Radiomics feature robustness across different CT devices and acquisition conditions is critical for the clinical translation of quantitative imaging biomarkers. This study aimed to systematically evaluate the robustness of radiomics features derived from photon-counting detector CT (PCD-CT) and three dual-energy CT (DECT) systems using chest tumor phantoms under different reconstruction images and scanning conditions.

Materials/Methods: Twelve chest tumor phantoms with varying sizes and texture characteristics were constructed using a CRIS chest phantom to simulate heterogeneous thoracic tumors. All phantoms were scanned on four energy CT devices: PCD-CT, dual-layer DECT (DL-DECT), dual-source DECT (DS-DECT), and rapid kV-switching DECT (RS-DECT). Images were reconstructed as conventional images (CI), virtual monoenergetic images (VMI) at 40, 70, and 100 keV, and virtual non-contrast (VNC) images. Tumor-specific regions of interest were individually delineated. After standardized image preprocessing, radiomics features were extracted. Inter-scanner robustness was evaluated using the concordance correlation coefficient (CCC), coefficient of variation (CV), and quartile coefficient of dispersion (QCD).

Results: Radiomics feature robustness across devices varied substantially depending on reconstruction types. In CI, for 1-mm and 3-mm spherical phantoms, the highest robustness was observed between DS-DECT and RS-DECT (CCC = 0.898 and 0.968, respectively), whereas CCC values between other device pairs were generally below 0.75. Among ten texture phantoms, robust agreement (CCC > 0.75) was observed primarily between DS-DECT and RS-DECT, and additionally between PCD-CT and DL-DECT. In VNC images, radiomics feature robustness between PCD-CT and the three DECT systems was consistently poor (CCC < 0.75), while robustness among the three DECT devices remained acceptable (CCC > 0.60). For VMI at 40 keV, CCC values between PCD-CT and DL-DECT or DS-DECT exceeded 0.75 for texture phantoms, whereas agreement with RS-DECT remained limited. Radiomic robustness at 70 and 100 keV was lower than that observed at 40 keV. Similar robustness patterns were observed across different radiation dose levels.

Conclusion: Radiomics feature robustness across energy CT platforms is highly dependent on reconstruction type. Compared with CI, VNC images demonstrated improved inter-DECT robustness; however, robustness between PCD-CT and DECT systems was reduced. Low-energy VMI substantially improved inter-scanner robustness, including for PCD-CT. Intra-scanner robustness across reconstruction types varied among platforms, with PCD-CT appearing more sensitive to tumor texture variations. These findings highlight the importance of reconstruction selection and standardization when applying radiomics across heterogeneous CT technologies.