Detectability of Small Low-Attenuation Lesions With Deep Learning CT Image Reconstruction: A 24-Reader Phantom Study.

Researchers

Journal

Modalities

Models

Abstract

Background: Iterative Reconstruction (IR) techniques are susceptible to contrast-dependent spatial resolution, limiting overall radiation dose reduction potential. Deep learning imaging reconstruction (DLIR) may mitigate this limitation. Objective: To evaluate low-contrast detectability performance and radiation saving potential of a DLIR algorithm in comparison with filtered back projection (FBP) and IR using a human multi-reader non-inferiority study design and task-based observer modeling. Methods: A dual-phantom construct, consisting of a low-contrast detectability module (21 low-contrast hypoattenuating objects in seven sizes [2.4-10 mm] and three contrast levels [-15, -10, -5 HU] embedded within liver-equivalent background) and a Catphan phantom, was imaged at five radiation exposures (CTDIvol range, 1.4-14.0 mGy; size-specific dose estimate, 2.5-25.0 mGy; 90%-, 70%-, 50%-, and 30%-reduced, and full radiation levels) using a MDCT scanner. Images were reconstructed using FBP, hybrid IR (ASiR-V), and DLIR (TrueFidelity). Twenty-four readers of varying experience levels evaluated images using a two-alternative forced choice. A task-based observer model (detectability index, d’) was calculated. Reader performance was estimated by calculating the AUC using a non-inferiority method. Results: Compared to FBP and IR methods at routine radiation levels, DLIR medium and high settings showed non-inferior performance through a 90% radiation reduction (except for DLIR medium setting at 70% reduced level). The IR method was non-inferior to routine radiation FBP only for 30% and 50% radiation reductions. No significant difference in d’ was observed between routine radiation FBP and DLIR high setting through a 70% radiation reduction. Reader experience was not correlated with diagnostic accuracy (R2=.005). Conclusion: Compared to FBP or IR methods at routine radiation levels, certain DLIR algorithm weightings yielded non-inferior low-contrast detectability with radiation reductions up to 90% as measured by 24 human readers, and up to 70% as assessed by a task-based observer model. Clinical Impact: DLIR has substantial potential to preserve contrast-dependent spatial resolution for the detection of hypoattenuating lesions at decreased radiation levels in a phantom model, addressing a major shortcoming of current IR techniques.