Abstract
Purpose To evaluate how deep learning image reconstruction (DLIR) affects diagnostic image quality and noise characteristics at different radiation dose levels and how it influences dose optimization compared with conventional reconstruction methods.
Method A systematic review was conducted using major databases for studies published between 2020 and 2026. Eligible studies assessed DLIR across multiple dose levels, reported radiation dose, included quantitative image quality metrics, and compared DLIR with filtered back projection or iterative reconstruction.
Results Thirty‑seven studies met criteria, and 15 were synthesized. Three themes emerged: (a) DLIR reduced noise by 15–82% and often improved signal-to-noise and contrast-to-noise ratios while preserving spatial resolution; (b) dose reduction potential ranged from 20–90%, with some regions achieving 73–97%; and (c) heterogeneity in protocols and metrics limited consistency, although DLIR generally maintained diagnostic performance.
Discussion DLIR improved image quality by adapting to image content, reducing noise, and preserving spatial resolution across dose levels, supporting meaningful dose‑reduction opportunities. However, variation in study design, reconstruction settings, and reporting limits the ability to define reliable dose‑reduction thresholds, and the absence of prospective diagnostic accuracy studies restricts clinical certainty. Quantitative metrics may also overestimate performance relative to human observers, indicating that although DLIR shows promise, clinical adoption requires more rigorous validation and standardized evaluation methods.
Conclusion DLIR may support substantial radiation dose reductions without compromising diagnostic quality. Stronger prospective evidence and more consistent evaluation standards are needed before aggressive dose‑reduction strategies can be widely implemented.
DOI
https://doi.org/10.52519/00256
Graduation Date
Summer 8-7-2026
Document Type
Thesis
Degree Name
Master of Science in Radiologic Science
Program
The School of Health Professions
Faculty Advisor
Kevin R. Clark
Committee Member
Jessyca B. Wagner
Director, Graduate Program
William A. Undie
Dean
Kimberly Hoggatt Krumwiede
Recommended Citation
Alberto, M. R. AI-Enhanced CT Reconstruction: Effects on Image Quality Across Dose Levels and Implications for Radiation Dose Optimization: A Systematic Review. [thesis]. University of Texas MD Anderson Cancer Center; 2026. https://doi.org/10.52519/00256
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 International License.
