Abdominal Dual-Energy CT with Deep Learning Reconstruction: Preserving Image Quality and Hepatic Lesion Conspicuity...

The objective was to evaluate the image quality and hepatic lesion conspicuity in a dual-low-dose (radiation and contrast volume) upper abdominal dual-energy CT (DECT) protocol utilizing deep learning image reconstruction (DLIR). After propensity score matching, 92 matched pairs underwent either chest DECT with upper abdominal coverage (LD group) or abdominal DECT (SD group). LD 50-keV images were reconstructed using adaptive statistical iterative reconstruction (LD-50-AR), DLIR-middle (LD-50-DM), and DLIR-high (LD-50-DH) while additional 60-keV images were reconstructed using DLIR-high (LD-60-DH). SD 60-keV images were reconstructed with AR (SD-60-AR). Objective metrics including noise, signal-to-noise (SNR), contrast-to-noise ratio (CNR), and beam-hardening artifact (BHA); subjective scores on overall image quality, diagnostic confidence, anatomic clarity, artifacts, and image noise were evaluated. The noise dose efficiency index (lower values indicate higher dose image quality transfer efficiency) was derived to assess the trade-off between image quality and radiation dose. For low-attenuation liver lesions, lesion-to-liver CNR (LLR) and subjective lesion conspicuity scores were assessed. Compared with SD group, the CTDIvol and contrast volume in the LD group were reduced by 33.27% and 25%. For the LD group, LD-50-DH images showed the highest image quality with 38.1-46.4% lower noise than LD-50-AR, comparable to the SD-60-AR; CNR increased 85.3-105.6%, BHA comparable to SD-60-AR and the highest subjective scores. The low-attenuation hepatic lesions showed good conspicuity. Dual-low-dose abdominal DECT utilizing 50 keV and DLIR-high achieved substantial radiation and contrast volume reductions while maintaining comparable noise, lesion conspicuity, and superior image quality relative to SD protocols with 60 keV and AR.