Return To Abstract Listing

Chest Imaging

E3319. Model-Based Iterative Reconstruction in Screening Low-Dose CT: Comparison with AIDR and FBP

Prosper A,  Ronco T,  Dissanayake S,  Hassani C,  Lee C. University of Southern California, Los Angeles, CA

Address correspondence to A. Prosper (aprosper@gmail.com)

Objective: The purpose of this study was to compare the image quality and nodule detection capability of a novel forward-projected model-based iterative reconstruction (FIRST) algorithm with adaptive iterative dose reduction (AIDR 3D) and filtered back-projection (FBP) reconstructions of low-dose CT (LDCT) scans performed for lung cancer screening.

Materials and Methods: Raw image data from 50 patients who underwent screening LDCT were reconstructed using FBP, AIDR 3D, and FIRST. CT scanning parameters met American Association of Physicists in Medicine published protocols. Objective image noise was measured in the descending aorta on body setting images and lung parenchyma on lung setting images for both FIRST and AIDR 3D reconstructions. Subjective image quality was compared between all three reconstructions in a blinded and randomized manner using a 5-point Likert scale by two thoracic radiologists. Pulmonary nodules were counted, measured, and characterized (solid vs part-solid vs ground-glass) on FIRST and AIDR 3D images, and results were compared. Wilcoxon signed-rank test was used to compare continuous outcomes between FIRST and AIDR. Kappa coefficient was used to assess the agreement in nodule characterization between FIRST and AIDR and between readers. Intraclass correlation was used to assess the agreement in continuous measurements including Likert scale.

Results: A total of 50 patients (19 women, 31 men) were included, with mean age of 60 years (range, 50–82 years) and median BMI of 28.3 kg/m2. Median CTDIvol was 1.6 mGy, and median dose-length product was 58.8 mGy/cm. Median image noise on body setting was 10.4 HU with FIRST and 15.7 HU with AIDR (p < 0.001). Median image noise on lung setting was 40.2 HU with FIRST and 72.4 HU with AIDR (p < 0.001). Subjective image quality of both FIRST and AIDR was considered much better than that of FBP by both readers. Subjective image quality assessment of FIRST compared with AIDR differed between the two readers. However, there were no significant differences in pulmonary nodule count, measurement, and characterization between FIRST and AIDR 3D for either reader.

Conclusion: In the context of lung cancer screening, FIRST reconstructed images have significantly decreased image noise compared to AIDR 3D. Though the two readers rated subjective image quality of FIRST compared to AIDR 3D differently, objective metrics of nodule count, measurement, and characterization were comparable. Compared to AIDR 3D, FIRST results in decreased image noise while maintaining nodule detection and characterization capability. FIRST may allow for further decreases in radiation dose for screening LDCT.