Ultra-High-Resolution Photon-Counting-Detector CT with a Dedicated Denoising Convolutional Neural Network for Enhanced Temporal Bone Imaging

Image resolution versus noise in PCD-CT. As image resolution increases, image noise also increases, which can limit the utility of high-resolution settings in PCD-CT.

The overall workflow of the proposed deep CNN denoising method. All training data originated from patient image series reconstructed using 2 iterative reconstruction strengths, QIR1 and QIR3, with thin-slice (0.2-mm) and thick-slice (0.4-mm) thicknesses, respectively. A multiple-slice input strategy was implemented to enhance the performance of the CNN.

Noise textures, NPS, and MTFc for the bone insert from axial slices of the ACR phantom for Hr84-QIR3, Hr96-QIR3, and Hr96-CNN, displayed under a fixed window and level.

Representative images of the modiolus, stapes footplate, and incudomallear joint using 3 different reconstructions: Hr84-QIR3, Hr96-QIR3, and Hr96-CNN (W/L: 4000/1000 HU). Enhanced visualization with improved resolution and reduced noise is demonstrated, as indicated by the yellow arrows, the Hr96-CNN provides improved delineation of three key anatomic structures. Image noise was quantified by measuring the SD of CT numbers within a circular ROI placed in a uniform soft-tissue area, with values recorded in the lower left corner of each image.

Rankings from 2 readers regarding overall image quality and delineation of 3 key anatomic structures. For all 3 structures and overall image quality, CNN-Hr96 images rank the highest. Dull purple indicates the first rank; medium gray, the second rank; gold, the third rank.