Image texture, low contrast liver lesion detectability and impact on dose: Deep learning algorithm compared to partial model-based iterative reconstruction

D Racine,H.G Brat,B Dufour,J.M Steity,M Hussenot,B Rizk,D Fournier,F Zanca

doi:10.1016/j.ejrad.2021.109808

Abstract

ObjectivesTo compare deep learning (True Fidelity, TF) and partial model based Iterative Reconstruction (ASiR-V) algorithm for image texture, low contrast lesion detectability and potential dose reduction. MethodsAnthropomorphic phantoms (mimicking non-overweight and overweight patient), containing lesions of 6 mm in diameter with 20HU contrast, were scanned at five different dose levels (2,6,10,15,20 mGy) on a CT system, using clinical routine protocols for liver lesion detection. Images were reconstructed using ASiR-V 0% (surrogate for FBP), 60 % and TF at low, medium and high strength. Noise texture was characterized by computing a normalized Noise Power Spectrum filtered by an eye filter. The similarity against FBP texture was evaluated using peak frequency difference (PFD) and root mean square deviation (RMSD). Low contrast detectability was assessed using a channelized Hotelling observer and the area under the ROC curve (AUC) was used as figure of merit. Potential dose reduction was calculated to obtain the same AUC for TF and ASiR-V. ResultsFBP-like noise texture was more preserved with TF (PFD from -0.043mm-1 to -0.09mm-1, RMSD from 0.12mm-1 to 0.21mm-1) than with ASiR-V (PFD equal to 0.12 mm-1, RMSD equal to 0.53mm-1), resulting in a sharper image. AUC was always higher with TF than ASIR-V. In average, TF compared to ASiR-V, enabled a radiation dose reduction potential of 7%, 25 % and 33 % for low, medium and high strength respectively. ConclusionCompared to ASIR-V, TF at high strength does not impact noise texture and maintains low contrast liver lesions detectability at significant lower dose.

Highlights

Technological developments in computed tomography (CT) have transformed patient care for many diseases and clinical scenarios
Change in noise texture is more pronounced with adaptive statistical iterative reconstruction-V (ASiR-V) (0.53 mm− 1 for root mean square deviation (RMSD) and 0.12 mm− 1 for peak frequency difference (PFD))
- First, for each algorithm, using a bootstrap method, a curve fit in terms of area under the ROC curve (AUC) as a function of radiation dose (CTDIvol) with a shape function of a*x^c / (x^c + b) was carried out from data provided by the Channelized Hotelling Observer (CHO) model observer (a, b and c are the parameters of the function and x represented CT dose index (CTDIvol))

Summary

Introduction

Technological developments in computed tomography (CT) have transformed patient care for many diseases and clinical scenarios. CT contributes to most of the population radiation exposure from medical X-ray imaging; for Abbreviations: ALARA, As Low As Reasonably Achievable; ASiR-V, adaptive statistical iterative reconstruction-V; AUC, area under the ROC curve; BMI, Body Mass Index; CHO, Channelized Hotelling Observer; CI, confidence interval; CNR, contrast-to-noise ratio; CT, computed tomography; CTDIvol, CT dose index; DDoG, difference of Gaussian; DLIR, deep learning iterative reconstruction; IR, iterative reconstruction; nNPSe, normalized noise power spectrum (NPS) with an eye filter; NI, Noise Index; PFD, peak frequency difference; RMSD, root mean square deviation; TF, True Fidelity. IR algorithms change noise texture and noise amplitude, resulting in a false high-quality visual impression with potentially massive dose reduction, especially if detectability of small low contrast objects are not taken into account [12]

Objectives

Methods

Results

Conclusion