Effect of different reconstruction algorithms on computer-aided diagnosis (CAD) performance in ultra-low dose CT colonography.

Purpose: To assess the effect of different reconstruction algorithms on computer-aided diagnosis (CAD) performance in ultra-low-dose CT colonography (ULD CTC).

Materials And Methods: IRB approval and informed consents were obtained. Thirty prospectively enrolled patients underwent non-contrast CTC at 120 kVp/10 mAs in supine and 100 kVp/10 mAs in prone positions, followed by same-day colonoscopy. Images were reconstructed with filtered back projection (FBP), 80% adaptive statistical iterative reconstruction (ASIR80), and model-based iterative reconstruction (MBIR). A commercial CAD system was applied and per-polyp sensitivities and numbers of false-positives (FPs) were compared among algorithms.

Results: Mean effective radiation dose of CTC was 1.02 mSv. Of 101 polyps detected and removed by colonoscopy, 61 polyps were detected on supine and on prone CTC datasets on consensus unblinded review, resulting in 122 visible polyps (32 polyps<6 mm, 52 6-9.9 mm, and 38≥10 mm). Per-polyp sensitivity of CAD for all polyps was highest with MBIR (56/122, 45.9%), followed by ASIR80 (54/122, 44.3%) and FBP (43/122, 35.2%), with significant differences between FBP and IR algorithms (P<0.017). Per-polyp sensitivity for polyps≥10 mm was also higher with MBIR (25/38, 65.8%) and ASIR80 (24/38, 63.2%) than with FBP (20/38, 58.8%), albeit without statistical significance (P>0.017). Mean number of FPs was significantly different among algorithms (FBP, 1.4; ASIR, 2.1; MBIR, 2.4) (P=0.011).

Conclusion: Although the performance of stand-alone CAD for ULD CTC can be improved, IR algorithms, particularly MBIR, were shown to significantly increase the per-polyp sensitivity of CAD compared to FBP according to this study. Therefore, as ULD CTC only requires 1.02mSv, specific optimization of CAD for ULD CTC and IR algorithms is strongly warranted to make ULD CTC with CAD clinically viable.