Performance analysis of deep learning-based iterative reconstruction algorithms in the evaluation of lung nodules

YANG Minghao; HAN Yicheng; LIU Hongwu; HOU Qingfeng

Abstract：Objective Using the United Imaging uCT968 Spiral CT as a reference,the artificial intelligence iterative recon-struction(AIIR)algorithm and Filtered Back Projection(FBP)algorithm were used to reconstruct the images,and the lung nod-ules evaluation software was used to analyze the influence of AIIR on the performance of lung nodule evaluation.Methods A retrospective selection of 67 patients with ground-glass nodules(GGNs)in their lungs was made.These patients underwent chest CT scans at the Shandong Provincial Hospital Affiliated to Shandong First Medical University.An analysis of these patients'ini-tial CT scans was conducted,with all original case images being reconstructed using both the AIIR(experimental group)and FBP(control group).These two sets of images were then evaluated using both the InferVISION and uAI lung nodule assessment software.Their evaluation results,SNR,and CNR were compared and their effective radiation doses were counted.We also as-sessed the impact of image quality on the accuracy of AI analysis.Results The automated analysis from the AI lung nodule as-sessment software revealed that AIIR algorithm significantly increased the detection rate of GGNs compared to the FBP.Addition-ally,images reconstructed via AIIR demonstrated a significant improvement in both SNR and CNR,in the meanwhile,reduced radiation dose and enhanced image quality.Conclusion In clinical application,the AIIR reconstruction algorithm is beneficial to improve the diagnosis rate of lung ground-glass nodules.Moreover,the low-dose scanning of the AIIR algorithm can further minimize the iatrogenic radiation dose,protecting patients.