Related Experiment Video
Updated: Apr 25, 2026
06:33
Construction of a Preclinical Multimodality Phantom Using Tissue-mimicking Materials for Quality Assurance in Tumor Size Measurement
Published on: July 29, 2013
10.7K
Deep learning "super resolution" versus iterative reconstruction: Phantom-based image quality assessment
Stephanie M Leon1, Daniella Fabri2, Colin J Schaeffer3
1Department of Radiology, University of Florida, Gainesville, Florida, USA.
Journal of Applied Clinical Medical Physics
|April 24, 2026
Summary
Canon's Precise IQ Engine (PIQE) Cardiac, a deep learning reconstruction algorithm, significantly enhances CT image quality and spatial resolution compared to iterative reconstruction (IR). PIQE offers potential for dose reduction without compromising contrast-to-noise ratio, especially with its 1024 matrix version.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Radiology
- Computed Tomography (CT)
Background:
- Deep learning reconstruction (DLR) is emerging as a replacement for iterative reconstruction (IR) in CT imaging.
- DLR algorithms offer potential for noise reduction and "super resolution," enhancing spatial resolution by training on higher-resolution images.
- This study evaluates the image quality performance of a specific DLR "super resolution" algorithm.
Purpose of the Study:
- To quantitatively compare Canon's Precise IQ Engine (PIQE) Cardiac (a DLR algorithm) against the AIDR Enhanced 3D iterative reconstruction (IR) algorithm.
- To characterize the image quality improvements offered by the PIQE Cardiac "super resolution" algorithm.
Main Methods:
- Quantitative phantom-based image quality measurements were performed using the ACR Gammex 464 phantom.
- Scans were acquired at varying dose levels (CTDIvol of 2.6, 15.7, and 32.9 mGy) with and without a torso ring.
- Image reconstructions included AIDR with a cardiac kernel and PIQE Cardiac with 512x512 and 1024x1024 matrices.
- Key metrics analyzed were task-based modulation transfer function (MTFtask), contrast-to-noise ratio (CNR), CNR for low-contrast objects (CNR_LO), normalized power spectrum (NPS), and noise magnitude.
Main Results:
- PIQE demonstrated improved MTFtask over AIDR across most frequencies, with PIQE 1024 showing superior performance to PIQE 512.
- AIDR exhibited a noise magnitude at least 44% higher than PIQE across all scenarios.
- While PIQE generally yielded higher CNR_LO values than AIDR, AIDR outperformed PIQE for smaller low-contrast objects at medium and high doses.
- PIQE enabled dose reduction in the small phantom without a loss in CNR_LO for the 25-mm disk.
Conclusions:
- PIQE significantly improves CT image quality compared to AIDR, based on quantitative phantom measurements.
- The PIQE algorithm presents opportunities for radiation dose reduction in CT examinations.
- PIQE 1024 enhances spatial resolution compared to PIQE 512 without introducing apparent trade-offs in image quality.
