Renal calculi are a common and often recurrent urinary disease with an incidence rate of 10–15%, which is increasing year by year.Many stone patients require multiple follow-up examinations to monitor changes in stone size, which helps determine the appropriate timing and method for treatment. However, the radiation dose from repeated examinations is a concern. Deep learning algorithms can help improve image quality while reducing radiation dose, and may be applied in stone imaging to reduce the radiation burden on patients.

A phantom which inserted the high contrast objective and low contrast objective with 2mm, 5mm and 8mm in size was scanned in low-dose and ultra-low dose radiation levels. Low dose images were reconstructed with Adaptive Iterative Dose Reduction 3D algorithm (LD-AIDR). Ultra-low dose images were reconstructed with filtered back projection algorithm (ULD-FBP), Adaptive Iterative Dose Reduction 3D algorithm (ULD-AIDR), Forward-projected model-based Iterative Reconstruction SoluTion algorithm (ULD-FIRST), AICE (ULD-AICE) and sharp AICE (ULD-AICEs). The noise power spectrum (NPS), task transfer function (TTF), and detectability were measured. Clinical patients with suspected urolithiasis were prospectively enrolled and underwent low dose CT, followed by ULDCT if any urinary stone was observed. The clinical images were reconstruction as same as phantom. The radiation dose, stone and other lesions characteristics, objective and subjective image quality were evaluated.