ObjectiveTo evaluate the feasibility of various strength levels of deep learning image reconstruction （DLIR） algorithms for improving non-contrast abdominal CT image quality at ultra-low radiation doses， by comparing ultra-low-dose DLIR images with low-dose filtered back projection （FBP） images. MethodsA prospective collection of 85 patients undergoing non-contrast abdominal CT scans was performed， and a self-controlled study method was employed to conduct low-dose （LD） group and ultra-low-dose （ULD） group scans. The LD group used a noise index of 10 and employed FBP for image reconstruction （LD-FBP group）. The ULD group used a noise index of 30 and employed DLIR at different levels （low， medium， high）， resulting in three subgroups of reconstructed images： ULD-DLIR-L， ULD-DLIR-M， and ULD-DLIR-H. For each group， CT values， standard devia-tion （SD）， signal-to-noise ratio （SNR）， and contrast-to-noise ratio （CNR） were measured and calculated for the liver， spleen， kidneys， aorta， psoas major， and subcutaneous fat. Effective dose （ED） was also recorded. Two radiologists independently performed subjective evaluations of image quality using a 5-point scale. ResultsCompared with the LD-FBP group， the ULD-DLIR-L group showed significantly lower SNR and CNR values in the liver， spleen， kidneys， aorta， and psoas major （P<0.001）， while the ULD-DLIR-H group exhibited significantly higher values （P<0.001）. The difference of SNR and CNR values for the ULD-DLIR-M group showed no statistically significant difference. For subjective evaluation， the scores of the ULD-DLIR-L and ULD-DLIR-M groups were lower than those of the LD-FBP group， while there was no statistically significant difference in scores between the ULD-DLIR-H group and the LD-FBP group. The ED value of the ULD group was approximately 88% lower than that of the LD group. ConclusionCompared with the LD-FBP group， the ULD-DLIR-H group significantly reduces SD values while increasing SNR and CNR values， effectively improving the image quality of non-contrast abdominal CT scans.

Objective:To explore the feasibility of applying deep learning image reconstruction (DLIR) in low-radiation dose liver dual-energy CT to further improve image quality, diagnostic confidence of lesion, and accuracy of iodine concentration (IC) measurement.Methods:This prospective cohort study enrolled 60 patients scheduled for enhanced liver CT at the First Affiliated Hospital of Anhui Medical University from June 2023 to January 2024. The participants were randomly assigned into the standard dose group and low radiation dose group with 30 cases in each using randomized block method. The standard radiation dose group underwent standard-radiation dose 120 kVp scans during the venous phase, while the low radiation dose group underwent low radiation dose scans with a rapid kVp-switching spectral scanning mode at 80 kVp and 140 kVp. The effective radiation dose (ED) was calculated for both groups. The standard radiation dose group was reconstructed using adaptive statistical iterative reconstruction-V (ASIR-V) algorithm 40% (AR40 120 kVp). The low radiation dose group using high-intensity DLIR (DLIR-H) to reconstructed 40 keV and 50 keV virtual monoenergetic images (VMI) (DH-VMI 40 keV, DH-VMI 50 keV). The image quality of the above three groups was objectively evaluated through the measurement of image noise and calculation of contrast-to-noise ratio (CNR) and signal-to-noise ratio (SNR) for the liver and portal vein; and the image quality was subjectively scored for image noise, contrast, lesion conspicuity, and diagnostic confidence. In the low radiation dose group, DLIR-H and ASIR-V40% reconstructed iodine maps were used to measure the liver and portal vein of IC values, standard deviations (SD), and coefficients of variation (CV). One-way analysis of variance or Kruskal-Wallis H test was used to compare the differences of subjective and objective image quality among the three groups, and paired t-test was used to compare the differences in measurement indexes between DLIR-H and ASIR-V40% reconstructed iodine maps. Results:The ED in the low radiation dose group [(2.2±0.5) mSv] was reduced by 56.8% compared to the conventional radiation dose group [(5.4±1.4) mSv]. Objective evaluations demonstrated that DH-VMI 40 keV had higher image noise, CNR, and SNR for liver and portal veins compared to AR40 120 kVp ( P<0.001). DH-VMI 50 keV had lower image noise ( P=0.200), with higher CNR and SNR for the liver and portal vein compared to AR40 120 kVp( P<0.001). In subjective evaluation, there was no statistically significant difference in image noise scores between DH-VMI 40 keV and AR40 120 kVp ( P>0.05), while the image noise score for DH-VMI 50 keV was lower than that of AR40 120 kVp ( P<0.05). Both DH-VMI 40 keV and DH-VMI 50 keV had higher scores for contrast, lesion conspicuity, and diagnostic confidence compared to those of AR40 120 kVp ( P<0.05). In the low radiation dose group, there was no statistically significant difference in IC values for the liver and portal vein between the ASIR-V40% and DLIR-H algorithm reconstructed iodine maps ( P>0.05). The SD and CV of liver and portal vein in the DLIR-H reconstructed iodine maps were lower than those in the ASIR-V40% reconstructed iodine maps ( P<0.001). Conclusions:DLIR can effectively reduce the image noise of low-energy (40, 50 keV) VMI, enhance lesion conspicuity and diagnostic confidence, and improve measurement accuracy without affecting IC values.

Objective:To explore the feasibility of applying deep learning image reconstruction (DLIR) in low-radiation dose liver dual-energy CT to further improve image quality, diagnostic confidence of lesion, and accuracy of iodine concentration (IC) measurement.Methods:This prospective cohort study enrolled 60 patients scheduled for enhanced liver CT at the First Affiliated Hospital of Anhui Medical University from June 2023 to January 2024. The participants were randomly assigned into the standard dose group and low radiation dose group with 30 cases in each using randomized block method. The standard radiation dose group underwent standard-radiation dose 120 kVp scans during the venous phase, while the low radiation dose group underwent low radiation dose scans with a rapid kVp-switching spectral scanning mode at 80 kVp and 140 kVp. The effective radiation dose (ED) was calculated for both groups. The standard radiation dose group was reconstructed using adaptive statistical iterative reconstruction-V (ASIR-V) algorithm 40% (AR40 120 kVp). The low radiation dose group using high-intensity DLIR (DLIR-H) to reconstructed 40 keV and 50 keV virtual monoenergetic images (VMI) (DH-VMI 40 keV, DH-VMI 50 keV). The image quality of the above three groups was objectively evaluated through the measurement of image noise and calculation of contrast-to-noise ratio (CNR) and signal-to-noise ratio (SNR) for the liver and portal vein; and the image quality was subjectively scored for image noise, contrast, lesion conspicuity, and diagnostic confidence. In the low radiation dose group, DLIR-H and ASIR-V40% reconstructed iodine maps were used to measure the liver and portal vein of IC values, standard deviations (SD), and coefficients of variation (CV). One-way analysis of variance or Kruskal-Wallis H test was used to compare the differences of subjective and objective image quality among the three groups, and paired t-test was used to compare the differences in measurement indexes between DLIR-H and ASIR-V40% reconstructed iodine maps. Results:The ED in the low radiation dose group [(2.2±0.5) mSv] was reduced by 56.8% compared to the conventional radiation dose group [(5.4±1.4) mSv]. Objective evaluations demonstrated that DH-VMI 40 keV had higher image noise, CNR, and SNR for liver and portal veins compared to AR40 120 kVp ( P<0.001). DH-VMI 50 keV had lower image noise ( P=0.200), with higher CNR and SNR for the liver and portal vein compared to AR40 120 kVp( P<0.001). In subjective evaluation, there was no statistically significant difference in image noise scores between DH-VMI 40 keV and AR40 120 kVp ( P>0.05), while the image noise score for DH-VMI 50 keV was lower than that of AR40 120 kVp ( P<0.05). Both DH-VMI 40 keV and DH-VMI 50 keV had higher scores for contrast, lesion conspicuity, and diagnostic confidence compared to those of AR40 120 kVp ( P<0.05). In the low radiation dose group, there was no statistically significant difference in IC values for the liver and portal vein between the ASIR-V40% and DLIR-H algorithm reconstructed iodine maps ( P>0.05). The SD and CV of liver and portal vein in the DLIR-H reconstructed iodine maps were lower than those in the ASIR-V40% reconstructed iodine maps ( P<0.001). Conclusions:DLIR can effectively reduce the image noise of low-energy (40, 50 keV) VMI, enhance lesion conspicuity and diagnostic confidence, and improve measurement accuracy without affecting IC values.

Muscle is one of the most important tissues of human body, which is distributed around various organs and bones. Skeletal muscle plays an important role in human activities and its functional changes are closely related to its own morphological structure. The study of the relationship between musculoskeletal structure and function can help us to understand the physiology basics of force and to guide clinical practices. Ultrasonography has been widely used in the research of muscle properties since it is real-time, fast, nonradiative and inexpensive. In recent years, there emerges various researches on image processing method for musculoskeletal ultrasonography, especially for dynamic ultrasonography. This paper presents a brief overview of the existing methods and key steps of ultrasound image processing of musculoskeletal.
Humans ; Image Processing, Computer-Assisted ; Muscle, Skeletal ; diagnostic imaging ; Ultrasonography