Detection of hypodense hepatic and renal lesions on abdominal CT reconstructed with deep learning image reconstruction technique in patients with large body habitus: A multi-reader study.
Authors
Affiliations (6)
Affiliations (6)
- Department of Radiology, Massachusetts General Hospital, Boston, USA.
- Department of Radiology, King Chulalongkorn Memorial Hospital, Bangkok, Thailand.
- Institute for Diagnostic and Interventional Radiology, University of Cologne, Cologne, Germany.
- Department of Radiology, University of Pittsburgh Medical Center, Pittsburgh, USA.
- Department of Radiology, Brown University, Providence, USA.
- Department of Radiology, Massachusetts General Hospital, Boston, USA. [email protected].
Abstract
To evaluate detectability of hypodense hepatic and renal lesions on portal venous phase CT reconstructed with deep learning image reconstruction (DLIR) versus adaptive statistical iterative reconstruction-V at 40% strength (ASIR-V 40) in patients with large body habitus. This single-center retrospective study included patients ≥ 90 kg who underwent abdominal CT (February- May 2021). Four reconstructions were generated: ASIR-V 40, and DLIR low (L), medium (M), and high (H) strengths. Six radiologists (three trainees, three experienced) evaluated image quality (IQ), diagnostic confidence, lesion conspicuity of up to two hypodense lesions per patient, and subcentimeter lesion visibility using a five-point Likert scale. Lesion-to-background contrast-to-noise ratio (CNR) was measured. Wilcoxon signed-rank and Chi-square tests were performed. Inter-reader agreement was calculated using Kendall's W. Among the 46 patients (mean weight 117 ± 14 kg), 86 lesions (38 liver, 48 renal) were analyzed. DLIR (all strengths) outperformed ASIR-V 40 in IQ, with DLIR-H yielding highest scores for liver (4.04 ± 0.70 vs. 3.77 ± 0.68) and renal lesions (4.25 ± 0.72 vs. 3.75 ± 0.66) (p < 0.05). For liver lesions, DLIR-H improved diagnostic confidence. For renal lesions, DLIR-H improved lesion conspicuity and visibility of subcentimeter lesions. Inter-reader agreement was excellent among experienced readers and good among trainees. Quantitatively, DLIR-M and DLIR-H showed higher CNR than ASIR-V 40 for liver (6.30, 8.62 vs. 4.98) and renal lesions (13.62, 17.79 vs. 11.71) (p < 0.001). In patients with large body habitus, DLIR-M and -H improved IQ and CNR with DLIR-H providing superior lesion conspicuity and diagnostic confidence; this study represents one of the first evaluations of DLIR performance in this specific patient population.