Evaluation of the Physical Properties of Deep Learning–based Noise Reduction Processing by Intensity in the Field of Simple X-ray Imaging

PURPOSE: Advanced intelligent Clear-IQ Engine (AiCE) is a deep learning-based noise reduction technique that has been reported to reduce image noise while preserving anatomical structures compared with conventional methods. However, the influence of different processing strengths on image quality characteristics has not been sufficiently evaluated under clinical conditions. The purpose of this study was to clarify the effect of AiCE processing strength on image quality in chest radiography. METHODS: Spatial resolution characteristics were evaluated using clinical exposure conditions for posteroanterior chest radiography at our institution. Images were acquired using a chest phantom and an aluminum cylinder, and the task transfer function (TTF) was calculated using the circular edge method. Noise characteristics were evaluated by acquiring images of a polymethyl methacrylate phantom and calculating the normalized noise power spectrum (NNPS). In both evaluations, analyses were performed while varying the AiCE processing strength (1, 5, 10, off). RESULTS: The TTF exhibited similar shapes across the entire spatial frequency range even when the AiCE processing strength was varied, and no consistent trend attributable to differences in processing strength was observed. In contrast, the NNPS decreased with the application of AiCE and showed a stepwise reduction with increasing processing strength. CONCLUSION: Under conditions of chest radiography, increasing the AiCE processing strength effectively reduced image noise while having minimal impact on spatial resolution characteristics.