Deep learning image reconstruction algorithm for abdominal multidetector CT at different tube voltages: assessment of image quality and radiation dose in a phantom study

Abstract

Objectives To compare the image quality and radiation dose of a deep learning image reconstruction (DLIR) algorithm compared with iterative reconstruction (IR) and filtered back projection (FBP) at different tube voltages and tube currents. Materials and methods A customized body phantom was scanned at different tube voltages (120, 100, and 80 kVp) with different tube currents (200, 100, and 60 mA). The CT datasets were reconstructed with FBP, hybrid IR (30% and 50%), and DLIR (low, medium, and high levels). The reference image was set as an image taken with FBP at 120 kVp/200 mA. The image noise, contrast-to-noise ratio (CNR), sharpness, artifacts, and overall image quality were assessed in each scan both qualitatively and quantitatively. The radiation dose was also evaluated with the volume CT dose index (CTDIvol) for each dose scan. Results In qualitative and quantitative analyses, compared with reference images, low-dose CT with DLIR significantly reduced the noise and artifacts and improved the overall image quality, even with decreased sharpness (p < 0.05). Despite the reduction of image sharpness, low-dose CT with DLIR could maintain the image quality comparable to routine-dose CT with FBP, especially when using the medium strength level. Conclusion The new DLIR algorithm reduced noise and artifacts and improved overall image quality, compared to FBP and hybrid IR. Despite reduced image sharpness in CT images of DLIR algorithms, low-dose CT with DLIR seems to have an overall greater potential for dose optimization.