【作者】 李伟；

【导师】 张建军；

【作者基本信息】 承德医学院 ， 影像医学与核医学（专业学位）， 2023， 硕士

【摘要】 目的:探讨常规MRI参数联合纹理分析鉴别鼻腔鼻窦内翻性乳头状瘤(SNIP)及其恶变的价值。方法:选取2016年10月至2021年6月于我院经手术及病理证实的41例SNIP患者和15例SNIP恶变患者。通过评价两组病变的临床资料、MRI征象及使用Ma Zda软件从MRI图像提取纹理特征参数,比较SNIP组与SNIP恶变组组间差异,并将获得的有统计学意义的指标进行二元logistic回归分析,建立MRI征象、纹理特征及MRI征象联合纹理特征预测模型,最后运用ROC曲线对各模型进行评价。结果:在SNIP组与恶变组的临床资料组间比较中,性别、年龄均无统计学差异(P>0.05)。在SNIP组与恶变组的MRI征象组间比较中,脑回征缺失、坏死、眼眶受累、颅脑受侵具有统计学意义(P<0.05),进行二元Logistic回归分析,颅脑受侵无统计学差异(P>0.05),脑回征缺失、坏死、眼眶受累是SNIP恶变的独立危险因素(P<0.05)。在SNIP组与恶变组的纹理特征组间比较中,S(1,0,0)Sum Entrp、S(0,1,0)Contrast、S(0,1,0)Dif Varnc、S(0,2,0)Contrast、S(0,2,0)Dif Varnc具有统计学意义(P<0.05),进行二元Logistic回归分析,最终筛选出一个纹理特征,S(0,2,0)Contrast是SNIP恶变的独立危险因素(P<0.05)。最后建立预测模型,所建立的MRI征象预测模型、纹理特征预测模型及联合预测模型的AUC值分别为0.865、0.793、0.924,且联合预测模型的敏感度、特异度及准确率均高于前两者(敏感度86.7%;特异度97.6%;准确率96.4%)。结论:常规MRI参数与纹理分析可以为SNIP及其恶变的鉴别提供客观依据,且采用MRI征象与纹理特征联合建模,其诊断效果明显优于仅采用单一模型的效果,为临床制定手术方案及评估预后提供更多信息。更多还原

【Abstract】 Objective:To analyze the value of conventional MRI signs combined with texture features in the differential diagnosis between sinonasal inverted papilloma and its malignant transformation.Methods:Forty-one patients with SNIP and 15 patients with SNIP malignant transformation confirmed by operation and pathology in our hospital from October 2016 to June 2021 were selected.By evaluating the clinical data,MRI signs and extracting texture feature parameters from MRI images by Ma Zda software of lesions in both groups,the differences between the SNIP group and the SNIP malignant transformation group were compared,and the statistically significant indexes were analyzed by binary logistic regression analysis,the prediction models of MRI signs,texture features and MRI signs combined with texture features were established,and finally,the ROC curve was used to evaluate each model.Results:In the group comparison of clinical data between the SNIP group and the malignant transformation group,there was no statistical difference in gender and age(P>0.05).In the group comparison of MRI signs between the SNIP group and the malignant transformation group,loss of convoluted cerebriform pattern,necrosis,orbit involvement and craniocerebral invasion were statistically significant(P<0.05),and when binary logistic regression analysis was performed,craniocerebral invasion was not statistically different(P > 0.05),and loss of convoluted cerebriform pattern,necrosis,orbit involvement were the independent risk factors for SNIP malignant transformation(P<0.05).In the comparison of texture features between the SNIP group and the malignant transformation group,S(1,0,0)Sum Entrp,S(0,1,0)Contrast,S(0,1,0)Dif Varnc,S(0,2,0)Contrast and S(0,2,0)Dif Varnc were statistically significant(P<0.05),a texture feature was screened out by binary Logistic regression,analysis S(0,2,0)Contrast was an independent risk factor for SNIP malignant transformation(P < 0.05).Finally,the prediction models were established,the AUC values of the MRI sign prediction model,texture feature prediction model and joint prediction model were 0.865,0.793,0.924,respectively.the sensitivity,specificity and accuracy of the joint prediction model were higher than the former two(sensitivity:86.7%;specificity:97.6%;accuracy:96.4%).Conclusion:Conventional MRI signs and texture analysis can provide an objective basis for the identification of SNIP and SNIP malignant transformation,in addition,the diagnostic efficacy of the model established by MRI signs combined with texture features is significantly better than that of the single model,which provides more information for clinical planning of surgery and evaluation of prognosis.更多还原