Image Quality Advancements in Low-Dose Pediatric CT Using Super-Resolution Deep-Learning Reconstruction

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Pediatric Radiology

Background:

• Low-dose CT scans are crucial for pediatric patients to minimize radiation exposure.
• Assessing image quality in low-dose pediatric CT is challenging due to inherent noise and reduced detail.
• Deep learning reconstruction (DLR) techniques offer potential for improving image quality in medical imaging.

Purpose of the Study:

• To evaluate the impact of super-resolution deep-learning reconstruction (SR-DLR) on image quality for low-dose, thin-slice pediatric abdominal CT.
• To compare SR-DLR with hybrid-iterative reconstruction (hybrid-IR) and conventional deep-learning reconstruction (C-DLR).

Main Methods:

• Retrospective analysis of low-radiation abdominal CT data from 38 children (under 10 years old).
• Generated 0.5-mm images using SR-DLR (1024 matrix), hybrid-IR (512 matrix), and C-DLR (512 matrix).
• Quantitative assessments included image noise (NPS), contrast-to-noise ratio (CNR), and edge-rise slope (ERS); qualitative assessments ranked noise, texture, sharpness, and structure delineation.

Main Results:

• SR-DLR demonstrated superior noise reduction and the highest CNR compared to hybrid-IR and C-DLR (p < 0.001).
• SR-DLR achieved higher edge sharpness (ERS) and maintained lower noise across spatial frequencies.
• Qualitative evaluations and diagnostic confidence scores were highest for SR-DLR across all metrics (p < 0.001).

Conclusions:

• SR-DLR with a 1024 matrix significantly enhances image sharpness and reduces noise in low-dose pediatric abdominal CT.
• SR-DLR outperforms hybrid-IR and C-DLR in delineating small structures and improving overall diagnostic quality.
• SR-DLR represents a promising advancement for pediatric abdominal CT imaging, balancing radiation dose reduction with diagnostic image quality.