基于深度学习重建的扩散加权成像在肺部病变良恶性鉴别中的应用价值

Application value of diffusion-weighted imaging reconstructed based on deep learning in benign and malignant differentiation of pulmonary lesions

目的评估深度学习重建(deep learning reconstruction,DLR)对肺部扩散加权成像(diffusion weighted imaging,DWI)图像质量的影响,并探讨DLR在肺部良恶性病变鉴别诊断中的价值。材料与方法前瞻性招募61例有肺内结节或肿块的患者(包括49例恶性和12例良性)。每位患者均使用3.0 T磁共振获得T2WI和DWI图像,DWI图像分别使用传统重建(conventional reconstruction,ConR)和DLR进行重建。两名具有4年和10年经验的放射科医生分别独立评估整体图像质量、病灶的信噪比(signal-to-noise ratio,SNR)、对比噪声比(contrast-to-noise ratio,CNR)和表观扩散系数(apparent diffusion coefficient,ADC)。采用Kappa值来评估测量者间在主观评分上的一致性,类相关系数(intra-class correlation coefficient,ICC)用于评估测量者间在SNR、CNR以及ADC值间的一致性。采用Wilcoxon秩和检验比较DLR DWI和ConR DWI分别在主观评分、SNR、CNR和ADC间的差异。采用Mann-Whitney检验比较良恶性病变间的差异。采用受试者工作特征(receiver operating characteristics,ROC)曲线评估ADC值鉴别病灶良恶性的诊断效能,并用DeLong检验比较ConR和DLR间曲线下面积(area under the curve,AUC)的差异。结果DLR和ConR的图像主观评分在观察者间均显示良好的一致性(Kappa>0.60)。客观评价中SNR和ADC均显示出极好的一致性(ICC>0.75),而CNR仅显示出一般的测量者间一致性(ICC>0.40)。与ConR DWI相比,DLR DWI具有更高的整体图像质量评分(P=0.003)、病变SNR(P<0.001)和更高的ADC值(P=0.017)。此外,DLR DWI的CNR也高于ConR DWI,但差异无统计学意义(P=0.258)。对ConR DWI和DLR DWI而言,恶性病变的ADC均显著低于良性病变(P<0.05)。ROC曲线分析显示DLR DWI(AUC=0.891)在良恶性病变中的诊断性能高于ConR DWI(AUC=0.808),DeLong检验显示差异具有统计学意义(P=0.044)。结论DLR DWI能够显著提高图像整体质量与图像SNR,相较于ConR DWI,DLR DWI能够显著提高鉴别肺部良恶性病变的诊断效能。

Objective:To evaluate the impact of deep learning reconstruction(DLR)on the image quality of pulmonary diffusion-weighted imaging(DWI)and to explore the value of DLR in the identification of benign and malignant pulmonary lesions.Materials and Methods:In this prospective study,61 patients with pulmonary nodules or masses(including 49 malignant and 12 benign cases)were recruited.Each patient underwent T2WI and DWI imaging using a 3.0 T MRI scanner,with DWI images reconstructed using both conventional reconstruction(ConR)and deep learning reconstruction(DLR).Two radiologists with 4 and 10 years of experience independently evaluated the overall image quality,signal-to-noise ratio(SNR),contrast-to-noise ratio(CNR),and apparent diffusion coefficient(ADC)of the lesion.Interobserver agreement on subjective scores was assessed using Kappa values,while intra-class correlation coefficient(ICC)were used to evaluate interobserver agreement on SNR,CNR,and ADC values.Wilcoxon rank sum tests were used to compare the differences between DLR DWI and ConR DWI in terms of subjective scores,SNR,CNR,and ADC.Mann-Whitney tests were performed to compare differences between benign and malignant lesions.The diagnostic performance of ADC for identifying benign and malignant lesions was evaluated using receiver operating characteristics(ROC)curves,with the area under the curve(AUC)compared between ConR and DLR using the DeLong test.Results:Both DLR and ConR images showed good interobserver agreement in subjective scoring(Kappa>0.60).In objective assessments,SNR and ADC demonstrated excellent interobserver consistency(ICC>0.75),while CNR showed only fair interobserver agreement(ICC>0.40).Compared to ConR DWI,DLR DWI had higher overall image quality scores(P=0.003),lesion SNR(P<0.001),and higher ADC values(P=0.017).Additionally,the CNR of DLR DWI was higher than that of ConR DWI,but the difference was not significant(P=0.258).For both ConR and DLR DWI,the ADC of malignant lesions was significantly lower than that of benign lesions(P<0.05).ROC curve analysis indicated that DLR DWI(AUC=0.891)had higher diagnostic performance in distinguishing between benign and malignant lesions compared to ConR DWI(AUC=0.808),with a significant difference observed by DeLong test(P=0.044).Conclusions:DLR DWI significantly improves overall image quality and enhances the SNR of images,offering superior diagnostic performance for distinguishing benign from malignant lesions compared to ConR DWI.