Ultra-high-resolution (UHR) photon-counting-detector (PCD) CT improves image resolution but increases noise, necessitating use of smoother reconstruction kernels that reduce resolution below the system's 0.110 mm maximum spatial resolution. To address this, a denoising convolutional neural network (CNN) was developed to reduce noise in images reconstructed with the available sharpest reconstruction kernel while preserving resolution for enhanced temporal bone visualization. With IRB approval, CNN was trained on 6 clinical temporal bone patient cases (1,885 images) and tested on 20 independent cases using a dual-source PCD-CT (NAEOTOM Alpha, Siemens). Images were reconstructed using iterative reconstruction at strength 3 (QIR3) with both clinical routine (Hr84) and the sharpest available head kernel (Hr96). The CNN was applied to images reconstructed with Hr96 and QIR1. Three image series (Hr84-QIR3, Hr96-QIR3, and Hr96-CNN) for each case were randomized for review by two neuroradiologists, assessing overall quality and delineation of the modiolus, stapes footplate, and incudomallear joint. CNN reduced noise by 80% compared to Hr96-QIR3 and 50% relative to Hr84-QIR3, while maintaining high resolution. When compared to the conventional method at the same kernel (Hr96-QIR3), Hr96-CNN significantly decreased image noise (from 204.63 HU to 47.35 HU) and improved SSIM (from 0.72 to 0.99). Hr96-CNN images ranked higher than Hr84-QIR3 and Hr96-QIR3 in overall quality (p<0.001). Readers preferred Hr96-CNN for all three structures. The proposed CNN significantly reduced image noise in UHR PCD-CT, enabling the use of sharpest kernel. This combination greatly enhanced diagnostic image quality and anatomical visualization.ABBREVIATIONS: PCD = Photon-counting-detector; UHR = Ultra-high-resolution; IR = Iterative reconstruction; CNN = Convolutional neural network; SSIM: Structural similarity index.
Read full abstract