Knowledge-based iterative model reconstruction (IMR) algorithm in ultralow-dose CT for evaluation of urolithiasis: evaluation of radiation dose reduction, image quality, and diagnostic performance.

PURPOSE: To evaluate the efficacy of a knowledge-based iterative model reconstruction (IMR) algorithm for reducing image noise in ultralow-dose (ULD) CT for urolithiasis.

MATERIALS AND METHODS: A total of 103 patients diagnosed with urinary stones (n = 276) were enrolled. Regular dose (RD) scans (120 kV and 150 mAs, maximal tube current in dose modulation) were reconstructed using filtered back-projection (FBP, RD-FBP), and ULD scans (100 kV and 20 mAs, fixed tube current) were reconstructed with FBP (ULD-FBP), statistical iterative reconstruction (IR; ULD-iDose), and a knowledge-based IMR algorithm (ULD-IMR). Prospective interpretations of the two scans were performed with respect to radiation dose, objective image noise, and subjective assessment. The subjective assessment was also evaluated with regard to each patient's body mass index (BMI, < 25 or ≥ 25 kg/m(2)). Using RD CT (RD-FBP) as the reference standard, two reviewers assessed the diagnostic performance and inter-observer agreement for ULD-IMR.

RESULT: The average effective doses with RD CT and ULD CT were 8.31 and 0.68 mSv, respectively, and the average radiation dose reduction rate was 91.82% (p < 0.01). The lowest objective image noise was observed with ULD-IMR (p < 0.01). Subjective assessment in ULD-IMR was comparable to that of RD-FBP, although RD-FBP remained statistically superior. For BMI, there was a statistically significant difference in subjective image quality between the normal (4.7 ± 0.54) and overweight or obese groups (4.2 ± 0.5) (p < 0.05). The ULD-IMR showed a greater than 75% concordant rate in overall stones and 100% in ureter stones larger than 3 mm. However, for stones <3 mm, neither reviewer had a good detection rate (45.5% and 56.9% for the general and genitourinary radiologist, respectively). Inter-observer agreement was almost perfect (κ = 0.82).

CONCLUSION: Despite a significant radiation dose reduction, ULD-IMR images were comparable in image quality and noise to RD-FBP images. Furthermore, the diagnostic performance of the ULD non-enhanced CT protocol was comparable to that of the RD scan for diagnosing urinary stones larger than 3 mm.