Improving Image Quality in Computed Tomography-Guided Biopsy Using Deep Learning Reconstruction

Computed tomography (CT)-guided biopsy offers procedural safety and is widely used for the diagnosis of lesions. Recently, CT systems with deep learning reconstruction (DLR) have become available, offering enhanced noise reduction and the potential to reduce radiation exposure. However, DLR does not support real-time imaging in CT fluoroscopy (CTF), limiting its use in interventional procedures. In this technical report, we investigate the image quality and reconstruction time of DLR in CT-guided biopsy using conventional (non-CTF) methods. Using a routine inspection phantom provided with the CT system, imaging was performed at setting standard deviation (SD) values of 20, 30, 40, and 50 Hounsfield units (HU); image noise and reconstruction time for three reconstruction methods (filtered back projection (FBP), hybrid iterative reconstruction (HIR), and DLR) were measured. SD was used as a quantitative measure of image noise. The results showed that DLR had the lowest image noise, followed by HIR and FBP, and the higher the setting SD, the more pronounced the difference. This indicates that DLR can improve image quality while maintaining diagnostic image quality and reducing radiation exposure. However, DLR had the longest reconstruction time, exceeding 10 seconds to output six images. Although DLR has obvious advantages in image quality and dose reduction, the associated delay in image reconstruction currently limits its use to procedures where real-time imaging is not essential. When real-time imaging is not required, DLR should be used in conjunction with conventional methods to reduce patient radiation exposure.