阈值校正方式对PET脑显像统计参数图分析显示结果的影响

Effects of the threshold correction on the results of statistic parameter mapping analysis of PET images

目的探讨各种阈值校正方法对统计参数图（SPM）软件统计比较结果显示的影响。方法利用Hoffman标准脑模型制作缺损模型PET成像与正常模型PET成像，进行统计参数图的统计比较。并选取脑梗死患者与健康检查者进行统计参数图的统计比较。结果校正与非校正产生的结果有差异，其中族错误率（FWE）校正（P=5×10^-2）得到的统计分析结果激活区最少及最小，其次是错误发现率（FDR）校正（P=5×10^-2），非校正方式（P=1×10^-3）得到的激活区最多及最大。但FDR校正效果不稳定，有时不如不校正。结论统计参数图软件中FWE校正方法可明显降低假阳性，得到的结果可信度稳定地高于非校正方法。

Objective To investigate the effects of the threshold correction on the results of the statistic parameter mapping （ SPM ） soft. Methods Defect model was established by a Hoffman standard brain model. The PET images of defect model were compared with the PET images of the standard modal by SPM. Images of the patients with cerebral infarction and healthy people were selected to conduct SPM statistical comparison. Results The correction results and non-correction results were different. Family wise error （FWE） correction had the more power and could get the least and the smallest regions, whose P value was higher than the threshold. The false discovery rate was unsteady, sometimes correction was worse than the non-correction. Conclusion In SPM, FWE correction has the least false region and the more reliable result than the non-correction.