Additional Dose Reduction Potential of Vendor-Agnostic Deep Learning Model: A Phantom Study.
Authors
Abstract
To evaluate the additional dose reduction potential (DRP) of a vendor-agnostic deep learning model (DLM, ClariCT.AI) applied to iterative reconstruction (IR) methods from two CT vendors. CT images of a multi-sized image quality phantom (Mercury v4.0) were obtained using two CT scanners from different vendors at six dose levels: 6%, 12%, 25%, 50%, 100%, and 200% of the reference dose (computerized tomography dose index volume; CTDIvol = 12 mGy). Images were reconstructed using three IR methods (ADMIRE, iDose<sup>4</sup>, IMR) at two strength levels and then denoised using the DLM. The detectability index (d') was measured using a 31 cm phantom across combinations of three target sizes (10, 5, 1 mm), five contrast levels (-895 to 1,000 Hounsfield unit), and six dose levels. DRP was defined as the percentage reduction in dose at which DLM-enhanced images achieved d' equivalent to that of reference-dose IR images. Mean DRPs were 93.2% for ADMIRE, 90.0% for iDose<sup>4</sup>, and 87.2% for IMR. Higher DRPs were observed for higher-contrast targets. Improvements in CNR was also noted in DLM-applied images. Applying a vendor-agnostic DLM to IR images enabled substantial radiation dose reduction while preserving image quality across different CT systems.