[Usefulness of Deep Learning Reconstruction in Low-dose Lung Cancer CT Screening Protocols].

Abstract

Purpose: This study aimed to evaluate the changes in physical characteristics when the image reconstruction method, radiation dose, and pitch factor (PF) were varied in low-dose lung cancer CT screening, and to determine the optimal radiation dose and PF for appropriate dose reduction and the usefulness of deep learning reconstruction (DLR).

Methods: Physical characteristics were evaluated using an Aquilion PrimeSP/i Edition (Canon Medical Systems, Tochigi) X-ray CT unit in conjunction with water phantoms and a chest phantom. Image reconstruction methods included filtered back projection (FBP), iterative reconstruction (IR), and DLR. Exposure conditions were varied across four dose levels and three PF levels. Physical characteristics were quantitatively evaluated using the noise power spectrum, task transfer function (TTF), low-contrast object-specific contrast-to-noise ratio (CNR_LO), and system performance function (SPF).

Results: Both the IR application method and DLR improved noise characteristics compared to FBP, even at low doses, and reduced noise in the high spatial frequency domain when the PF level was lowered. DLR improved TTF at low doses and SPF at a standard deviation (SD) of 50. There was no significant difference in CNR_LO by PF level.

Conclusion: DLR may be useful in low-dose lung cancer CT screening, and appropriate SD settings and PF selection may contribute to image optimization.