Impact of reconstruction algorithms on the success rate and quality of automatic airway segmentation in children under ultra-low-dose chest CT scanning

Background: To investigate the success rate and quality of automatic airway segmentation using ultra-low dose CT (ULD-CT) images of different reconstruction algorithms. Materials and Methods: Fifty two children who underwent chest ULD-CT were divided into three groups for analysis based on age: group A (n=13, age, 1-2years), group B (n=19, age, 3-6years) and group C (n=20, age, 7-13years). CT images were reconstructed with filtered back-projection (FBP), 50% adaptive statistical iterative reconstruction-Veo (50%ASIR-V), 100%ASIR-V, deep learning image reconstruction (DLIR) with low (DLIR-L), medium (DLIR-M), and high (DLIR-H) strengths. Subjective image quality was evaluated using a 5-point scale. CT value, noise, and sharpness of the trachea were measured. The VCAR software was used to automatically segment airways and reported the total volume. Segmentation success rates were recorded, and segmentation images were subjectively evaluated using a 6-point scale. Results: The average tracheal diameters were 8.53±1.88mm, 10.69±1.65mm, and 12.72±1.97mm, respectively for groups A, B, and C. The segmentation success rate depended on patient groups: group C reached 100%, while group A decreased significantly. In group A, 100%ASIR-V had the lowest rate at 7.69%, while DLIR-M and DLIR-H significantly improved the rate to 38.64% (P=0.03). For the segmented images, DLIR-H provided the lowest noise and highest subjective score while FBP images had the highest noise and 100%ASIR-V had the lowest overall score (P<0.05). There was no significant difference in the total airway volume among the six reconstructions. Conclusion: The airway segmentation success rate in ULD-CT for children depends on the tracheal size. DLIR improves airway segmentation success rate and image quality.