Advances in the diagnosis of acute abdomen: Low dose CT with deep learning image reconstruction is noninferior to standard dose CT; results from a prospective multicenter study

Purpose or Learning Objective
Contrast-enhanced abdominopelvic CT is commonly performed to assess acute abdomen; , albeit with a high radiation dose. Efforts have been made to reduce dose with iterative image reconstruction techniques; however deteriorating image quality has made clinical implementation difficult. Deep learning image reconstruction (DLR) technique may improve images to acceptable diagnostic quality.
Methods or Background
In this prospective multicenter, single-blinded, cross-sectional non-inferiority trial, adult patients with acute abdomen referred to contrast-enhanced CT were included. Each patient underwent two CT scans, one with a standard radiation dose (SDCT) immediately followed by a low dose scan (LDCT). SDCT was performed according to each hospital’s local standard. LDCT was reconstructed with DLR. A total of 18 readers with varied experience assessed the scans; diagnosis and reading time were recorded. The primary endpoint was the diagnostic accuracy of the LDCT compared to SDCT. The reference standard was the clinical radiology report from the SDCT. Secondary endpoints were reader experience, obesity (BMI>=30), and reading time. Mixed-effects logistic/linear regression was used as appropriate to evaluate associations between accuracy, radiation dose, and secondary outcomes.
Results or Findings
A total of 196 patients participated, with 149 showing acute pathological findings. The median CTDIvol of SDCT was 12.0 mGy (range 6.6-45.0 mGy), for LDCT 2.7 mGy (1.5-13.0 mGy).
The diagnostic accuracy of LDCT was 83.7% (95%CI 79.9%-87.4%), which was noninferior to SDCT with 82.9% (78.8-86.5%), p=0.807. Obesity (20) did not reduce accuracy, p=0.793. Reader experience of 5-10 years had a small negative effect, p=0.033.
LDCT did not increase reading time, p=0.420.
Conclusion
Low-dose contrast-enhanced abdominopelvic CT is accurate and feasible for diagnosis in the clinical setting of acute abdomen.
Limitations
Only two centers participated; the number of patients differed between centers (center 01 n=168; center 02 n=28).