CT纹理分析鉴别诊断血液病并发侵袭性肺曲霉病或肺毛霉菌病

CT texture analysis for differential diagnosis of hematologic disease combined invasive pulmonary aspergillus or pulmonary mucormycosis

目的观察CT纹理分析鉴别诊断血液病并发侵袭性肺曲霉菌病(IPA)或肺毛霉菌病(PM)的价值。方法回顾性分析111例血液病患者的胸部CT,其中65例合并IPA(IPA组),46例合并PM(PM组)。采用IBEX软件分析CT图像,手动勾画病灶ROI,并提取纹理特征参数,计算特征值;比较组间纹理特征值差异,选取差异有统计意义的纹理特征参数,采用受试者工作特征(ROC)曲线分析其鉴别诊断IPA与PM的效能。结果共于2组提取979个CT纹理特征参数,其中繁忙度(Busyness)、粗糙度(Coarseness)、纹理强度(TextureStrength)及表面密度(SurfaceAreaDensity)的特征值组间差异均有统计学意义。ROC曲线分析结果显示,TextureStrength的曲线下面积(AUC)最大(0.807),敏感度76.52%,特异度82.69%;多参数联合分析显示,联合应用4个参数的AUC最大(0.915),敏感度87.01%,特异度96.83%。结论CT纹理分析可用于鉴别血液病并发IPA或PM,多参数联合可提高其诊断效能。

Objective To observe the value of CT texture analysis for differential diagnosis of hematologic disease combined invasive pulmonary aspergillosis(IPA)or pulmonary mucormycosis(PM).Methods CT images of 111 patients with hematologic diseases were retrospectively analyzed,including 65 patients complicated with IPA(IPA group)and 46 with PM(PM group).IBEX software was used to manually delineate pulmonary lesions and extract their texture features.The extracted feature values were statistically analyzed,those being different between groups were selected,and receiver operating characteristic(ROC)curve was used to analyze the differential diagnosis efficiency of CT texture feature for IPA and PM.Results A total of 979 parameters were extracted from CT texture features of IPA and PM,among which Busyness,Coarseness,TextureStrength and SurfaceAreaDensity were statistically different between groups.ROC curve analysis results showed that TextureStrength had the highest diagnostic efficacy,with AUC of 0.807,sensitivity of 76.52%and specificity of 82.69%.The combined analysis of multiple parameters showed that with combining application of four parameters,AUC reached the maximum(0.915),with sensitivity of 87.01%and specificity of 96.83%.Conclusion CT texture analysis could be used to distinguish hematologic disease combined with IPA or PM,and combination of multiple parameters could improve the diagnostic efficiency.