Improved visualization of pancreas and tumor boundaries using high-frequency kernels with deep-learning image reconstruction at high-strength level.
Authors
Affiliations (8)
Affiliations (8)
- Department of Radiology, Gifu University, 1-1 Yanagido, Gifu, 501-1194, Japan.
- Department of Radiology, Gifu University, 1-1 Yanagido, Gifu, 501-1194, Japan. [email protected].
- Department of Frontier Science for Imaging, Gifu University, 1-1 Yanagido, Gifu, 501-1194, Japan. [email protected].
- Department of Radiology Services, Gifu University Hospital, 1-1 Yanagido, Gifu, 501-1194, Japan.
- Department of Frontier Science for Imaging, Gifu University, 1-1 Yanagido, Gifu, 501-1194, Japan.
- Department of Food Hygiene and Control, Faculty of Veterinary Medicine, Suez Canal University, Ismailia, Egypt.
- Innovation Research Center for Quantum Medicine, Graduate School of Medicine, Gifu University, 1-1 Yanagido, Gifu, 501-1194, Japan.
- Center for One Medicine Innovative Translational Research (COMIT), Institute for Advanced Study, Gifu University, 1-1 Yanagido, Gifu, 501-1194, Japan.
Abstract
This study aimed to investigate the feasibility of combining high-frequency reconstruction kernels and deep-learning image reconstruction at high-strength level (DLIR-H) for improving visualization of the pancreas and tumor boundaries on pancreatic protocol CT. This retrospective study included 30 patients (median age, 75 years; 16 women) who underwent pancreatic protocol CT for assessing pancreatic tumors from January 2024 to July 2024. Four image sets were reconstructed using DLIR-H in combination with either standard, bone, bone-plus, or lung kernels. Edge sharpness between the pancreas and retroperitoneal fat tissue (pancreas-to-fat) and between the pancreas and pancreatic ductal adenocarcinoma (pancreas-to-PDAC) was quantitatively assessed using edge rise slope (ERS) measurements. Two radiologists qualitatively examined the sharpness of the pancreas, tumor boundary, and overall image quality. Pancreas-to-fat ERS was greater in lung kernel images than in standard and bone kernel images (P = 0.001). Pancreas-to-PDAC ERS was greater in lung kernel images than in other kernel images (P < 0.001). Sharpness of the pancreas and tumor boundaries was better in lung kernel images than in other kernel images (P < 0.001 for both). Overall image quality in lung kernel images was comparable to the standard and superior to the bone kernel images (P < 0.001). The combination of lung kernel and DLIR-H in pancreatic protocol CT improves both quantitative and qualitative sharpness of the pancreas and tumor boundaries while maintaining the overall image quality.