Reducing artifacts of arms-down positioning in abdominal CT with artificial intelligence iterative reconstruction

Purpose: To test the feasibility of an artificial intelligence iterative reconstruction (AIIR) for reducing artifacts of arms-down positioning in abdominal CT. Methods and Materials: This study retrospectively enrolled 49 patients (11 females and 38 males; mean age of 55.0 ± 19.4 years) who were unable to raise their arms during abdominal CT scans. The raw data were reconstructed using three methods: the hybrid iterative reconstruction (HIR), the HIR combined with adaptive filtering (HIR+AF), and the AIIR. The signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) on the liver, kidney, pancreas, and spleen were measured and compared among the three image sets. The arm-artifacts, image contrast, diagnostic acceptability, overall image quality, and lesion sharpness were evaluated by two radiologists using a 5-point Likert scale ranging from 1 (poor) to 5 (excellent). Results: The SNRs of AIIR for the liver, kidney, pancreas, and spleen were 6.0 ± 1.7, 3.2 ± 1.1, 4.1 ± 1.4, and 5.4 ± 1.5, respectively, significantly higher than those of HIR (2.4 ± 1.1, 1.4 ± 0.6, 1.8 ± 0.7, and 2.1 ± 1.0) and HIR+AF (4.0 ± 1.2, 2.4 ± 0.9, 3.0 ± 0.9, and 3.7 ± 1.2) (all p < 0.01). Similarly, the CNRs of AIIR for the liver, kidney, pancreas, and spleen were 17.7 ± 3.9, 15.6 ± 3.3, 14.9 ± 3.8, and 18.0 ± 4.9, respectively, significantly higher than those of HIR (6.1 ± 2.4, 5.0 ± 1.9, 5.3 ± 1.9, and 5.7 ± 2.6) and HIR+AF (10.4 ± 2.7, 8.9 ± 2.4, 9.1 ± 2.6, and 10.2 ± 3.7) (all p < 0.01). Compared to HIR and HIR+AF, the AIIR showed the highest image quality scores regarding arm-artifacts, image contrast, diagnostic acceptability, overall image quality, and lesion sharpness (all p < 0.001). Conclusion: AIIR can significantly improve image quality for patients with arms-down positioning in abdominal CT, compared to HIR and HIR+AF.