The Application Of Deep Learning Reconstruction Algorithm In Low-Dose Contrast-enhanced Abdominal CT

Objective: To evaluate deep learning image reconstruction(DLIR)on image quality of low-dose contrast-enhanced abdominal CT,comparing to adaptive statistical iterative reconstruction(ASIR-V).Materials and Method: Fifty-two patients who needed contrast-enhanced abdominal CT were prospectively collected.Low dose CT was used in the arterial phase and conventional dose CT was used in the portal vein phase and delayed phase.Arterial-phase images were reconstructed using 40%ASIR-V,80%ASIR-V(1.25 mm and 5mm)(ASIRV was prepositioned)and DL-M,DL-H(1.25mm).Portal phase images were reconstructed using 40%ASIR-V(1.25 mm thickness).CT values of liver,aorta,spleen and erector spinal muscle were measured,and SD values of subcutaneous fat were measured as background noise.Contrast-to-noise ratio(CNR)was calculated as the result of image objective evaluation.Subjective evaluation of the images was performed by two radiologists in a double-blind method.Results :Compared with 40%ASIR-V,80%ASIR-V and DL-M in the same layer thickness at low dose,DL-H had the lowest background noise SD(8.63±2.35HU),the highest CNR and the higher subjective score(4.79±0.41).At low dose,the CNR of DL-H(1.25 mm)was lower than 80%ASIR-V(5mm),the CNR of DL-M(1.25 mm)was lower than 40%ASIR-V(5mm),but the subjective score of DLIR(1.25 mm)was greater than ASIR-V(5mm).The SD values of DL-H at low dose(8.63±2.35HU)were lower than the SD values of 40%ASIR-V at conventional dose(11.54±2.97HU).Conclusion: Compared with ASIR-V,DLIR can significantly reduce image noise and improve image quality,and has higher clinical value in Low-Dose Contrast-enhanced Abdominal CT.