Stability of Radiomics Features in Photon-Counting Detector CT: Acquisition and Reconstruction Effects
Overview
This phantom study evaluated the robustness of radiomics features extracted from photon-counting detector CT (PCD-CT) images under varying acquisition and reconstruction parameters. The findings highlight how factors such as scan mode, tube voltage, slice thickness, radiation dose, iterative reconstruction level, and reconstruction kernel influence radiomics feature stability.
Background
Radiomics translates medical images into quantitative data, showing promise in diagnosis, staging, prognosis, and treatment response assessment. Despite growing interest and methodological guidelines, radiomics models often lack generalizability and robustness, limiting clinical translation. Previous studies have demonstrated variability in radiomics features across conventional and dual-energy CT systems. Photon-counting detector CT (PCD-CT) is a novel technology with limited data on radiomics feature stability, particularly regarding the impact of acquisition and reconstruction parameters within the system.
Data Highlights
Parameter
Levels Tested
Scan Mode
Standard vs High-pitch
Tube Voltage
120 kVp vs 140 kVp
Slice Thickness
1.0 mm vs 0.4 mm
Radiation Dose
0.5, 1.0, 3.0, 5.0, 10.0 mGy
Quantum Iterative Reconstruction (QIR) Level
0/4, 2/4, 4/4
Reconstruction Kernel
Qr40, Qr44, Qr48
Key Findings
Radiomics feature values are sensitive to changes in acquisition parameters such as scan mode and tube voltage.
Reconstruction parameters including slice thickness, iterative reconstruction level, and reconstruction kernel significantly affect radiomics feature stability.
Repeatability and reproducibility of radiomics features were quantitatively assessed using intraclass correlation coefficient (ICC) and concordance correlation coefficient (CCC).
Variability metrics such as coefficient of variation (CV) and quartile coefficient of dispersion (QCD) were used to estimate feature variability across parameter changes.
The study utilized a homemade texture phantom with 28 different material inserts to simulate diverse textures for radiomics analysis.
Virtual monochromatic images at 70 keV were used for feature extraction to standardize comparisons.
Clinical Implications
Understanding the influence of acquisition and reconstruction parameters on radiomics feature stability in PCD-CT is critical for developing robust radiomics models. Standardizing imaging protocols and reconstruction settings may improve the reproducibility and clinical applicability of radiomics-based biomarkers derived from PCD-CT. Clinicians and researchers should consider these factors when designing radiomics studies or implementing radiomics in clinical practice.
Conclusion
This study demonstrates that acquisition and reconstruction variables significantly impact radiomics feature robustness in PCD-CT. Careful optimization and standardization of these parameters are essential to enhance the reliability and clinical translation of PCD-CT radiomics.
References
Zhao et al. 2024 -- Evaluating the Stability of Radiomics in Photon-Counting Detector CT
