Image quality improvement in low-dose chest CT with deep learning image reconstruction

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Low-dose computed tomography (CT) is essential for lung cancer screening.
• Image quality in low-dose CT can be compromised by noise.
• Advanced reconstruction algorithms are needed to enhance diagnostic accuracy.

Purpose of the Study:

• To evaluate the clinical utility of deep learning image reconstruction (DLIR).
• To compare DLIR with the adaptive statistical iterative reconstruction-Veo (ASiR-V40%) algorithm for low-dose chest CT.
• To assess DLIR's impact on image quality and noise reduction.

Main Methods:

• Retrospective analysis of 86 patients undergoing low-dose chest CT for lung cancer screening.
• Image reconstruction using ASiR-V40% and three DLIR levels (low, medium, high).
• Quantitative measurement of CT values and noise; subjective image quality assessment by two blinded readers.

Main Results:

• DLIR reconstructions significantly reduced image noise in the aorta compared to ASiR-V40% (up to 56.1% reduction).
• DLIR demonstrated significantly improved subjective image quality regarding artifacts (p < 0.001).
• Visualization of small structures was comparable between ASiR-V40%, DLIR-L, and DLIR-M, with DLIR-H showing a slight decrease.

Conclusions:

• Deep learning image reconstruction (DLIR) offers significant noise reduction benefits for low-dose chest CT.
• The strength of DLIR should be tailored to specific clinical diagnostic requirements.
• DLIR represents a valuable advancement for improving low-dose CT imaging.