Abstract

Objectives: This study investigates the repeatability and reproducibility of radiomic features extracted from different energy levels of virtual monoenergetic reconstruction (VMER) and polyenergetic reconstruction (PER) obtained with photon-counting computed tomography (PCCT). Methods: Sixteen organic phantoms were scanned twice in a test–retest format using a 120 kV tube potential and tube currents of 10, 50, and 100 mAs. After rotating the phantoms 90° around their z-axis, additional test–retest scans were performed. A PER and 16 VMERs were generated. Segmentation and extraction of 105 original radiomic features followed. The repeatability and reproducibility of these features were assessed using the concordance correlation coefficient (CCC) for agreement and the intraclass correlation coefficient (ICC) for reliability, excluding 14 shape-based features from the analysis. Results: On average, 85 out of 91 radiomic features from VMER showed high repeatability. Approximately 30% of features demonstrated high intra-scan and inter-scan reproducibility when comparing PER and VMER. For different energy levels of VMER, around 78% showed high intra-scan reproducibility, and 74% showed high inter-scan reproducibility. Comparing the average values of test and retest scans in both the initial and rotated states revealed that 65% of features showed high agreement and 73% high reliability for PER, while for VMER, these values were 51% and 55%, respectively. Conclusions: Radiomic features from VMERs showed high test–retest repeatability, whereas reproducibility across reconstruction types and widely separated energy levels was more limited. These findings suggest that energy levels should be carefully standardized when radiomic features are extracted from PCCT-derived VMER images.

IPC Classification

Keywords