Dependence of Resolution Properties on Display Field-of-view Size in Deep Learning-based Reconstruction Computed...

Deep learning-based reconstruction (DLR) has transformed computed tomography (CT) imaging by providing superior noise reduction and image enhancement compared to conventional filtered back projection (FBP). This study aimed to elucidate the impact of display field-of-view (D-FOV) size on the spatial resolution properties of DLR images, particularly when assessing small vascular signals relevant to head CT angiography.

A vascular phantom containing circular signals with diameters of 0.5 mm and 1.0 mm, filled with iodine contrast medium, was used. Images were reconstructed using three methods: FBP, model-based iterative reconstruction (IR), and DLR, all with standard denoising settings. Three D-FOV sizes (100 mm, 200 mm, and 400 mm) were evaluated. Spatial resolution was quantitatively assessed using task transfer function (TTF) measurements, derived from the Fourier transforms of the vascular signal profiles.

FBP images exhibited consistent TTF values regardless of D-FOV size. In contrast, both IR and DLR images showed lower TTF values for the smaller (0.5 mm) signals compared to the 1.0 mm signals across all D-FOV settings. Moreover, TTF values in IR and DLR images decreased progressively with increasing D-FOV size.

This study demonstrates that spatial resolution, as quantified by TTF, is adversely affected by both smaller signal diameters and larger D-FOV sizes in IR and DLR reconstructions.