[Noise Reduction Effect of Deep-learning-based Image Reconstruction Algorithms in Thin-section Chest CT].

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Thin-section chest CT imaging is crucial for diagnosing pulmonary conditions.
• Image noise can degrade diagnostic accuracy and image quality.
• Traditional reconstruction methods like filtered back projection (FBP) and adaptive statistical iterative reconstruction (ASIR) have limitations in noise reduction.

Purpose of the Study:

• To compare the noise reduction capabilities of deep learning image reconstruction (DLIR) algorithms against conventional methods (FBP, ASIR) in thin-section chest CT.
• To evaluate the impact of DLIR on image quality metrics and subjective assessment.

Main Methods:

• Chest CT raw data from 47 patients were analyzed.
• Images were reconstructed using FBP, ASIR (50%, 70%), and DLIR (low, medium, high modes).
• Quantitative analysis of CT values, standard deviation (SD), and signal-to-noise ratio (SNR) was performed on selected regions of interest (aorta, muscle, lung).
• Two radiologists assessed overall image quality.

Main Results:

• Statistically significant differences (P<0.001) were observed in CT values, SD, and SNR across all six reconstruction methods.
• Image quality scores also showed significant differences (P<0.001) between methods.
• The DLIR high mode (DL-H) yielded the lowest noise levels and the highest image quality scores.

Conclusions:

• Deep learning-based reconstruction effectively reduces noise and improves image quality in thin-section chest CT.
• DLIR, particularly the DL-H mode, offers superior noise reduction compared to FBP and ASIR.
• DLIR holds significant promise for enhancing diagnostic performance in chest CT imaging.