Task-based characterization of a deep learning image reconstruction and comparison with filtered back-projection and a partial model-based iterative reconstruction in abdominal CT: A phantom study

Damien Racine; Fabio Becce; Anais Viry; Pascal Monnin; Brian Thomsen; Francis R Verdun; David C Rotzinger

doi:10.1016/j.ejmp.2020.06.004

Task-based characterization of a deep learning image reconstruction and comparison with filtered back-projection and a partial model-based iterative reconstruction in abdominal CT: A phantom study

Phys Med. 2020 Aug:76:28-37. doi: 10.1016/j.ejmp.2020.06.004. Epub 2020 Jun 20.

Authors

Damien Racine¹, Fabio Becce², Anais Viry¹, Pascal Monnin¹, Brian Thomsen³, Francis R Verdun¹, David C Rotzinger⁴

Affiliations

¹ Institute of Radiation Physics (IRA), Lausanne University Hospital (CHUV) and University of Lausanne (UNIL), Rue du Grand-Pré 1, 1007 Lausanne, Switzerland.
² Department of Diagnostic and Interventional Radiology, Lausanne University Hospital (CHUV) and University of Lausanne (UNIL), Rue du Bugnon 46, 1011 Lausanne, Switzerland.
³ GE Healthcare, Milwaukee, WI, USA.
⁴ Department of Diagnostic and Interventional Radiology, Lausanne University Hospital (CHUV) and University of Lausanne (UNIL), Rue du Bugnon 46, 1011 Lausanne, Switzerland. Electronic address: david.rotzinger@chuv.ch.

PMID: 32574999
DOI: 10.1016/j.ejmp.2020.06.004

Abstract

Purpose: We aimed to thoroughly characterize image quality of a novel deep learning image reconstruction (DLIR), and investigate its potential for dose reduction in abdominal CT in comparison with filtered back-projection (FBP) and a partial model-based iterative reconstruction (ASiR-V).

Methods: We scanned a phantom at three dose levels: regular (7 mGy), low (3 mGy) and ultra-low (1 mGy). Images were reconstructed using DLIR (low, medium and high levels) and ASiR-V (0% = FBP, 50% and 100%). Noise and contrast-dependent spatial resolution were characterized by computing noise power spectra and target transfer functions, respectively. Detectability indexes of simulated acute appendicitis or colonic diverticulitis (low contrast), and calcium-containing urinary stones (high contrast) (|ΔHU| = 50 and 500, respectively) were calculated using the nonprewhitening with eye filter model observer.

Results: At all dose levels, increasing DLIR and ASiR-V levels both markedly decreased noise magnitude compared with FBP, with DLIR low and medium maintaining noise texture overall. For both low- and high-contrast spatial resolution, DLIR not only maintained, but even slightly enhanced spatial resolution in comparison with FBP across all dose levels. Conversely, increasing ASiR-V impaired low-contrast spatial resolution compared with FBP. Overall, DLIR outperformed ASiR-V in all simulated clinical scenarios. For both low- and high-contrast diagnostic tasks, increasing DLIR substantially enhanced detectability at any dose and contrast levels for any simulated lesion size.

Conclusions: Unlike ASiR-V, DLIR substantially reduces noise while maintaining noise texture and slightly enhancing spatial resolution overall. DLIR outperforms ASiR-V by enabling higher detectability of both low- and high-contrast simulated abdominal lesions across all investigated dose levels.

Keywords: Computed tomography; Deep learning; Image quality; Iterative reconstruction; Model observer; Radiation dose.

MeSH terms

Algorithms
Deep Learning*
Image Processing, Computer-Assisted
Radiation Dosage
Radiographic Image Interpretation, Computer-Assisted*
Tomography, X-Ray Computed