CT radiomic features reproducibility of virtual non-contrast series derived from photon-counting CCTA datasets using a novel calcium-preserving reconstruction algorithm compared with standard non-contrast series: focusing on epicardial adipose tissue.

Purpose: We aimed to evaluate the reproducibility of computed tomography (CT) radiomic features (RFs) about Epicardial Adipose Tissue (EAT). The features derived from coronary photon-counting computed tomography (PCCT) angiography datasets using the PureCalcium (VNC) and conventional virtual non-contrast (VNC) algorithm were compared with true non-contrast (TNC) series.

Methods: RFs of EAT from 52 patients who underwent PCCT were quantified using VNC, VNC, and TNC series. The agreement of EAT volume (EATV) and EAT density (EATD) was evaluated using Pearson's correlation coefficient and Bland-Altman analysis. A total of 1530 RFs were included. They are divided into 17 feature categories, each containing 90 RFs. The intraclass correlation coefficients (ICCs) and concordance correlation coefficients (CCCs) were calculated to assess the reproducibility of RFs. The cutoff value considered indicative of reproducible features was > 0.75.

Results: the VNC and VNC tended to underestimate EATVs and overestimate EATDs. Both EATV and EATD of VNC series showed higher correlation and agreement with TNC than VNC series. All types of RFs from VNC series showed greater reproducibility than VNC series. Across all image filters, the Square filter exhibited the highest level of reproducibility (ICC = 67/90, 74.4%; CCC = 67/90, 74.4%). GLDM_GrayLevelNonUniformity feature had the highest reproducibility in the original image (ICC = 0.957, CCC = 0.958), exhibiting a high degree of reproducibility across all image filters.

Conclusion: The accuracy evaluation of EATV and EATD and the reproducibility of RFs from VNC series make it an excellent substitute for TNC series exceeding VNC series.