Impacts of Adaptive Statistical Iterative Reconstruction-V and Deep Learning Image Reconstruction Algorithms on Robustness of CT Radiomics Features: Opportunity for Minimizing Radiomics Variability Among Scans of Different Dose Levels

• DOI: 10.1007/s10278-023-00901-1
• Publication Date: 2024-01-29T18:02:35Z
• Authors: Jingyu Zhong; Zhiyuan Wu; Lingyun Wang; Yong Chen; Yihan Xia; Lan Wang; Jianying Li; Wei Lu; Xiaomeng Shi; Jianxing Feng; Haipeng Dong; Huan Zhang; Weiwu Yao
• Keywords: Concordance correlation coefficient; Radon transform

Abstract This study aims to investigate the influence of adaptive statistical iterative reconstruction-V (ASIR-V) and deep learning image reconstruction (DLIR) on CT radiomics feature robustness. A standardized phantom was scanned under single-energy (SECT) dual-energy (DECT) modes at standard low (20 10 mGy) dose levels. Images SECT 120 kVp corresponding DECT kVp-like virtual monochromatic images were generated with filtered back-projection (FBP), ASIR-V 40% (AV-40) 100% (AV-100) blending levels, DLIR algorithm (DLIR-L), medium (DLIR-M), high (DLIR-H) strength Ninety-four features extracted via Pyradiomics. Reproducibility calculated between algorithms in reference FBP images, within scan mode, using intraclass correlation coefficient (ICC) concordance (CCC). The average percentage ICC > 0.90 CCC two levels 21.28% 20.75% AV-40 39.90% 35.11% AV-100 respectively, increased from 15.43 45.22% 44.15% an increasing level DLIR. 26.07% 25.80% 18.88% 18.62% decreased 27.93 17.82% 27.66 17.29% showed reproducibility while high-strength provides opportunity for minimizing variability due reduction.