目的随着深度伪造技术的快速发展,人脸伪造图像越来越难以鉴别,对人们的日常生活和社会稳定造成了潜在的安全威胁。尽管当前很多方法在域内测试中取得了令人满意的性能表现,但在检测未知伪造类型时效果不佳。鉴于伪造人脸图像的伪造区...目的随着深度伪造技术的快速发展,人脸伪造图像越来越难以鉴别,对人们的日常生活和社会稳定造成了潜在的安全威胁。尽管当前很多方法在域内测试中取得了令人满意的性能表现,但在检测未知伪造类型时效果不佳。鉴于伪造人脸图像的伪造区域和非伪造区域具有不一致的源域特征,提出一种基于多级特征全局一致性的人脸深度伪造检测方法。方法使用人脸结构破除模块加强模型对局部细节和轻微异常信息的关注。采用多级特征融合模块使主干网络不同层级的特征进行交互学习,充分挖掘每个层级特征蕴含的伪造信息。使用全局一致性模块引导模型更好地提取伪造区域的特征表示,最终实现对人脸图像的精确分类。结果在两个数据集上进行实验。在域内实验中,本文方法的各项指标均优于目前先进的检测方法,在高质量和低质量FaceForensics++数据集上,AUC(area under the curve)分别达到99.02%和90.06%。在泛化实验中,本文的多项评价指标相比目前主流的伪造检测方法均占优。此外,消融实验进一步验证了模型的每个模块的有效性。结论本文方法可以较准确地对深度伪造人脸进行检测,具有优越的泛化性能,能够作为应对当前人脸伪造威胁的一种有效检测手段。展开更多
Objective To compare the value and consistency among the Patient Generated-Subjective Global Assessment(PG-SGA)and the Prognostic Nutrition Index(PNI)for assessing nutritional status in gastrointestinal tumor patients...Objective To compare the value and consistency among the Patient Generated-Subjective Global Assessment(PG-SGA)and the Prognostic Nutrition Index(PNI)for assessing nutritional status in gastrointestinal tumor patients.Methods 251 patients from gastric cancer surgical ward from January 2019 to January 2020 were recruited through convenience sampling in this respective study.Nutritional screening and assessment were conducted for 251 gastrointestinal tumor patients using the nutritional risk screening 2002(NRS 2002)PG-SGA,and the PNI.PNI was calculated using the serum albumin level and the total lymphocyte count obtained from the patients’routine laboratory examination when they were admitted to the hospital.The receiver operating characteristic(ROC)of the PG SGA and the PNI were plotted with the NRS 2002 used as the gold standard,and the diagnostic value of the PG-SGA and PNI was reflected by the area under the curve(AUC),sensitivity,specificity and Youden index.We then determined the optimal cut-off for the PNI and tested the consistency of the PG-SGA and PNI.Results The optimal cut-off point for the PNI was calculated to be 50.78.The AUC of the PG-SGA was 0.908(95%CI 0.871-0.944).The sensitivity was 89.9%,specificity was 76.2%and the Youden index was 0.661.The AUC of the PNI was 0.594(95%CI 0.516-0.572).The sensitivity was 73.8%,specificity was 44.3%and the Youden index was 0.181.In the consistency test,the kappa value was 0.838(P<0.001).Conclusion The PNI is of limited value for assessing malnutrition,although it did have good consistency with the PG-SGA.The combination of the PNI and PG-SGA can be used for diagnosing assessing malnutrition in clinical practice.展开更多
文摘目的随着深度伪造技术的快速发展,人脸伪造图像越来越难以鉴别,对人们的日常生活和社会稳定造成了潜在的安全威胁。尽管当前很多方法在域内测试中取得了令人满意的性能表现,但在检测未知伪造类型时效果不佳。鉴于伪造人脸图像的伪造区域和非伪造区域具有不一致的源域特征,提出一种基于多级特征全局一致性的人脸深度伪造检测方法。方法使用人脸结构破除模块加强模型对局部细节和轻微异常信息的关注。采用多级特征融合模块使主干网络不同层级的特征进行交互学习,充分挖掘每个层级特征蕴含的伪造信息。使用全局一致性模块引导模型更好地提取伪造区域的特征表示,最终实现对人脸图像的精确分类。结果在两个数据集上进行实验。在域内实验中,本文方法的各项指标均优于目前先进的检测方法,在高质量和低质量FaceForensics++数据集上,AUC(area under the curve)分别达到99.02%和90.06%。在泛化实验中,本文的多项评价指标相比目前主流的伪造检测方法均占优。此外,消融实验进一步验证了模型的每个模块的有效性。结论本文方法可以较准确地对深度伪造人脸进行检测,具有优越的泛化性能,能够作为应对当前人脸伪造威胁的一种有效检测手段。
文摘Objective To compare the value and consistency among the Patient Generated-Subjective Global Assessment(PG-SGA)and the Prognostic Nutrition Index(PNI)for assessing nutritional status in gastrointestinal tumor patients.Methods 251 patients from gastric cancer surgical ward from January 2019 to January 2020 were recruited through convenience sampling in this respective study.Nutritional screening and assessment were conducted for 251 gastrointestinal tumor patients using the nutritional risk screening 2002(NRS 2002)PG-SGA,and the PNI.PNI was calculated using the serum albumin level and the total lymphocyte count obtained from the patients’routine laboratory examination when they were admitted to the hospital.The receiver operating characteristic(ROC)of the PG SGA and the PNI were plotted with the NRS 2002 used as the gold standard,and the diagnostic value of the PG-SGA and PNI was reflected by the area under the curve(AUC),sensitivity,specificity and Youden index.We then determined the optimal cut-off for the PNI and tested the consistency of the PG-SGA and PNI.Results The optimal cut-off point for the PNI was calculated to be 50.78.The AUC of the PG-SGA was 0.908(95%CI 0.871-0.944).The sensitivity was 89.9%,specificity was 76.2%and the Youden index was 0.661.The AUC of the PNI was 0.594(95%CI 0.516-0.572).The sensitivity was 73.8%,specificity was 44.3%and the Youden index was 0.181.In the consistency test,the kappa value was 0.838(P<0.001).Conclusion The PNI is of limited value for assessing malnutrition,although it did have good consistency with the PG-SGA.The combination of the PNI and PG-SGA can be used for diagnosing assessing malnutrition in clinical practice.
