Application Value of Deep Learning Reconstruction in Low-dose Chest CT for COPD

Objective: To compare image quality between deep learning image reconstruction (DLIR) low-dose chest CT and iterative reconstruction (ASIR-V) standard-dose chest CT in patients with chronic obstructive pulmonary disease (COPD). Methods: A total of 106 patients were prospectively enrolled and underwent standard-dose (SD) and low-dose (LD) chest CT scans. The LD scans were reconstructed using ASIR-V (LD-AR) and three DLIR strength levels (LD-DL/DM/DH), whereas the SD scans were reconstructed using ASIR-V (SD-AR). Noise (standard deviation), signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR) of the anatomical structures were measured or calculated. The subjective image quality was scored by radiologists. Results: Effective radiation dose was (4.0±1.37) mSv for the SD group and (1.14±0.47) mSv for the LD group. The LD-DLIR images exhibited lower noise, higher SNR, and higher CNR than the SD-AR images, with the LD-DH performing the best. Subjective scores indicated superior noise levels, anatomical clarity, and emphysema visualization in LD-DLIR images compared to SD-AR, with LD-DH receiving the highest score. Conclusion: DLIR significantly improved chest CT image quality in patients with COPD while reducing the radiation dose by 71.5%, with DLIR-H providing optimal performance.