Quantitative and qualitative assessment of ultra-low-dose paranasal sinus CT using deep learning image reconstruction: a comparison with hybrid iterative reconstruction

Chuluunbaatar Otgonbaatar ¹, Dakyong Lee ², Juneho Choi ²

⁰Medical Imaging AI Research Center, Canon Medical Systems Korea, Seoul, Republic of Korea.

European Radiology +

|

June 13, 2025

View abstract on PubMed

Summary

Deep learning image reconstruction (DLR) significantly improves ultra-low-dose computed tomography (CT) for paranasal sinus imaging compared to hybrid iterative reconstruction. This advanced technique enhances image quality and sharpness, enabling effective surgical planning with reduced radiation exposure.

Area Of Science

Radiology
Medical Imaging
Image Reconstruction

Background

Ultra-low-dose computed tomography (CT) is crucial for paranasal sinus (PNS) imaging, especially for patients needing repeated scans or functional endoscopic sinus surgery (FESS) planning.
Minimizing radiation exposure is a key concern in CT imaging, necessitating advanced reconstruction techniques.
Traditional iterative reconstruction (IR) methods may have limitations in preserving image quality at very low radiation doses.

Purpose Of The Study

To quantitatively and qualitatively evaluate deep learning image reconstruction (DLR) performance against hybrid iterative reconstruction (IR) for ultra-low-dose PNS CT.
To assess the impact of DLR on image noise, signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and image sharpness.
To determine the diagnostic efficacy of DLR for preoperative paranasal sinus imaging.

Main Methods

A retrospective analysis of 132 patients who underwent non-contrast ultra-low-dose sinus CT (0.03 mSv).
Image reconstruction was performed using both hybrid IR and DLR algorithms.
Objective metrics (noise, SNR, CNR, noise power spectrum, perceptual sharpness) and subjective radiologist evaluations were used to assess image quality.

Main Results

DLR demonstrated significantly lower image noise (28.62 ± 4.83 HU) compared to hybrid IR (140.70 ± 16.04 HU).
DLR yielded superior SNR (22.47 ± 5.82 vs 9.14 ± 2.45) and CNR (71.88 ± 14.03 vs 11.81 ± 1.50).
Images reconstructed with DLR were significantly sharper (0.56 ± 0.04) and rated higher in overall quality, bone visualization, and diagnostic confidence by radiologists.

Conclusions

Deep learning image reconstruction (DLR) significantly outperforms hybrid iterative reconstruction (IR) in ultra-low-dose paranasal sinus CT.
DLR effectively reduces image noise, enhances SNR and CNR, and improves image sharpness while maintaining anatomical detail.
Ultra-low-dose CT with DLR provides sufficient image quality for preoperative planning, minimizing radiation exposure and enhancing patient safety.

Keywords:

Computed tomography Image quality Image reconstruction Paranasal sinuses