The application of deep learning-based artificial intelligence algorithms combined with low-dose scanning protocols in chest CT.

Background: Under low-dose scanning conditions, different reconstruction algorithms have varying effects on image quality. This study aimed to investigate the effects of the precise imaging (PI) deep-learning artificial intelligence (AI) algorithm combined with a low-dose scanning protocol on image quality and radiation dose in chest computed tomography (CT) scans. Methods: A retrospective analysis was conducted of 100 patients who underwent non-contrast chest CT scans at Union Hospital, Tongji Medical College, Huazhong University of Science and Technology, using Philips Incisive CT between October and December 31, 2024. The patients were divided into two groups: the experimental group (n=50) and the control group (n=50). The experimental group was scanned using a low-dose protocol with a tube voltage of 100 kVp, and the images were reconstructed using the PI deep-learning AI algorithm at a high-intensity level (Group A) and conventional iDose iterative reconstruction (Group B) for both the mediastinal and lung window settings, respectively. The control group (Group C) was scanned using a conventional-dose protocol with a tube voltage of 120 kVp, and the images were reconstructed by iDose iterative reconstruction. Objective image quality metrics, including the mean CT value, standard deviation (SD), signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR) of the regions of interest (ROIs) in the axial images, were measured and calculated. The statistical analysis was performed using SPSS software. Additionally, two experienced radiologists independently evaluated the image quality subjectively using a five-point scale (on which 1 represented poor, and 5 represented excellent). Results: The noise level (SD value) of the mediastinal and lung window images in experimental Group A was significantly lower than that in Group B (P<0.01). Additionally, the SNR and CNR values of the images were significantly higher in Group A than Group B (P<0.01). No significant differences were observed in the SD, CNR, and SNR values between experimental Group A and control Group C (P>0.05). The subjective image quality scores of Groups A, B, and C were 4.40±0.53, 1.42±0.50, and 4.30±0.51, respectively, with no significant difference between Groups A and C (P>0.05). For the experimental group (Groups A and B) and control Group C, the mean volume CT dose index (CTDIvol) values were 1.37±0.22 and 7.07±1.70 mGy, the mean dose-length product (DLP) values were 56.36±9.82 and 296.8±80.72 mGy·cm, and the mean effective dose (ED) values were 0.79±0.14 and 4.16±1.13 mSv, respectively. Compared to the control group, the CTDIvol, DLP, and ED in the experimental group were reduced by 81.00%. Conclusions: The low-dose scanning protocol combined with high-intensity PI deep learning-based algorithm reconstruction achieves high image quality while maintaining a low radiation dose, making it suitable for widespread application in chest CT screening.