深度学习重建算法改善腹部门静脉期CT图像质量的应用价值

The application value of deep learning imaging reconstruction algorithms for improving the quality of abdominal portal venous phase CT images

目的:探究深度学习重建算法(Deep learning image reconstruction,DLIR)、传统滤波反投影(Filered back-projection,FBP)及自适应迭代重建算法(Adaptive statistical iterative reconstruction-veo,ASIR-V)对改善腹部门静脉期CT图像质量差异及临床获益。方法:前瞻性纳入45例行腹部增强CT扫描患者,其中包括18例肝硬化失代偿期患者,对门静脉期图像进行FBP、30%ASIR-V、80%ASIR-V及DLIR-H重建,并测量比较4组重建图像肝脏、脾脏、脾静脉、门静脉及左右支CT值、噪声、信噪比(Signal-to-noise ratio,SNR)及对比信噪比(Contrast-to-noise ratio,CNR);比较各重建算法图像主观评价,包括18例肝硬化失代偿期患者交通支血管。结果:4组重建算法图像CT值无统计学差异(P>0.05),噪声、SNR、CNR均有统计学差异,两两比较FBP与30%ASIR-V,80%ASIR-V与DLIR-H在CNR、SNR值中无统计学差异(校正P<0.008),80%ASIR-V与DLIR-H算法在SD值无统计学差异(校正P<0.008),余均有统计学差异。主观评价DLIR图像整体质量、对比度、失真伪影与其他各组有显著性差异(校正P<0.008),仅图像噪声与80%ASIR-V无显著性差异(校正P≥0.008)。DLIR交通支血管轮廓、清晰度与各组有显著性差异(校正P<0.008),噪声与80%ASIR-V无显著性差异(校正P≥0.008)。结论:DLIR算法降低腹部CT图像噪声,改善图像质量具有优势,尤其是肝硬化失代偿期微小血管结构,该重建算法可能为患者的精准诊断、风险评估提供更多信息。

Objective:To compare the differences in image quality and clinical benefits of deep learning image reconstruction(DLIR),filtered back-projection(FBP),and adaptive statistical iterative reconstruction-veo(ASIR-V)in abdominal portal venous phase CT images.Methods:Forty-five patients who underwent abdominal contrast-enhanced CT scans were enrolled,and18 cases with decompensated liver cirrhosis were contained.The portal venous phase images were reestablished by FBP,30%ASIR-V,80%ASIR-V,and DLIR-H algorithms.The CT values and noise of the liver,spleen,splenic vein,portal vein,and left and right branches in each reconstructed image,as well as the signal-to-noise ratio(SNR)and contrast-to-noise ratio(CNR)were measured and compared.The subjective evaluations of each reconstructed image,including collateral vessels in 18 cases with decompensated liver cirrhosis.Results:There was no statistically significant difference in CT values among the four reconstructed image groups(P>0.05).However,there were statistically significant differences in noise,SNR,and CNR.Comparisons between FBP and 30%ASIR-V,as well as 80%ASIR-V and DLIR-H,showed no statistically significant differences in CNR and SNR values(adjusted P<0.008).There were no statistically significant differences in SD values between 80%ASIR-V and DLIR-H algorithms(adjusted P<0.008),but differences were observed in other comparisons.Subjective evaluation showed statistically significant differences in overall quality,contrast,and distortion/artifacts of DLIR images compared to other groups(adjusted P<0.008).Only image noise in DLIR did not show significant differences compared to 80%ASIR-V(adjusted P≥0.008).The delineation of vascular structures and clarity in DLIR images showed significant differences compared to other groups(adjusted P<0.008),with no significant differences in noise compared to 80%ASIR-V.Conclusion:The DLIR algorithm offers advantages in reducing noise and improving image quality of abdominal CT images,particularly in the visualization of small vascular structures in patients with decompensated liver cirrhosis.This reconstruction algorithm may potentially provide more information for accurates patients diagnosis and risk assessment.