Feasibility of radiation dose reduction with AI-based spectral reconstruction for computed tomography

Abstract Purpose The purpose of this study is to investigate whether 50% radiation dose with AI-based spectral reconstruction has equivalent or non-inferior image quality to routine dose with iterative reconstruction. Methods Fifteen patients were scanned at two radiation dose levels during a standard renal ablation procedure. Ablation planning images were obtained at routine dose (CTDI vol = 18.9 ± 4.7 mGy), and probe placement images were obtained at low dose (CTDI vol = 9.9 ± 2.5 mGy). Quantitative assessment included noise measurements and line profiles of iterative reconstruction (iDose, Philips) and AI-based spectral reconstruction (Spectral Precise Image, Philips) with respect to filtered-back-projection. Qualitative assessment included a radiologist reader study comparing routine dose images with iterative reconstruction (RD + IR) and low dose images with AI-based spectral reconstruction (LD + AI). Two radiologists rated overall image quality on a five-point Likert scale ranging from very poor (1) to excellent (5). A Wilcoxon signed-rank test was used to test for a significant difference between the reconstruction conditions. Results For conventional results, LD + AI was rated significantly higher than RD + IR ( p = 0.0051). For 50 keV MonoE and electron density results, there was not a significant difference between the LD and RD protocols. Of the results tested, the image quality of 50 keV was rated higher than conventional and electron density results. Conclusion This study suggests the feasibility of reducing radiation dose with AI-based spectral reconstruction without compromising overall image quality requirements for CT-guided renal ablations. For the conventional result, readers indicated superior image quality in LD + AI relative to RD + IR result.