ISSN 1004-4140
CN 11-3017/P
凡健, 高斌, 夏春华. CT平扫与动脉期图像纹理分析在鉴别膀胱乳头状瘤和膀胱癌中的应用价值

与恶性肿瘤的初步研究

[J]. CT理论与应用研究, 2020, 29(6): 742-750. DOI: 10.15953/j.1004-4140.2020.29.06.13
引用本文: 凡健, 高斌, 夏春华. CT平扫与动脉期图像纹理分析在鉴别膀胱乳头状瘤和膀胱癌中的应用价值

与恶性肿瘤的初步研究

[J]. CT理论与应用研究, 2020, 29(6): 742-750. DOI: 10.15953/j.1004-4140.2020.29.06.13
FAN Jian, GAO Bin, XIA Chunhua. The Value of CT Non-enhanced and Enhanced Image Texture Analysis in Differentiating Bladder Papilloma from Bladder Cancer[J]. CT Theory and Applications, 2020, 29(6): 742-750. DOI: 10.15953/j.1004-4140.2020.29.06.13
Citation: FAN Jian, GAO Bin, XIA Chunhua. The Value of CT Non-enhanced and Enhanced Image Texture Analysis in Differentiating Bladder Papilloma from Bladder Cancer[J]. CT Theory and Applications, 2020, 29(6): 742-750. DOI: 10.15953/j.1004-4140.2020.29.06.13

CT平扫与动脉期图像纹理分析在鉴别膀胱乳头状瘤和膀胱癌中的应用价值

与恶性肿瘤的初步研究

The Value of CT Non-enhanced and Enhanced Image Texture Analysis in Differentiating Bladder Papilloma from Bladder Cancer

  • 摘要: 目的:探讨利用CT平扫与动脉期图像纹理分析在鉴别膀胱乳头状瘤和膀胱癌中的应用价值。方法:回顾性纳入自2016年1月至2020年1月在安徽医科大学第三附属医院就诊的经病理证实的64例膀胱肿瘤患者的病例资料,其中良性病变32例,恶性病变32例。所有患者均行CT三期动态增强扫描,选择其平扫与动脉期图像来进行研究。使用MaZda纹理分析软件对扫描图像上的膀胱病变进行纹理特征的提取,再用费希尔参数(Fisher)法、最小分类误差与最小平均相关系数法(POE+ACC)及相关信息测度法(MI)分别筛选出鉴别膀胱良恶性病灶的10个最佳纹理参数,使用Mazda的B11工具的主成分分析法(PCA)、线性判别分析法(LDA)和非线性判别分析法(NDA)对选取的最佳纹理特征进行分析,计算出其鉴别膀胱良恶性肿瘤的最小误判率(R),对最小误判率所对应的最佳纹理参数进行ROC检验,筛选出最有辅助鉴别意义的可量化参数来对膀胱良恶性肿瘤进行鉴别诊断。结果:基于CT平扫时以MI+NDA组合的误判率最低(1.56%),所筛选的最佳纹理参数分别为WavEnHH_s-1(高高频小波系数转换s-1)、Horzl_Fraction (水平游程图像分数)、Horzl_ShrtREmp (水平短游程补偿)、Sigma (参数σ)、Variance (方差),对应的AUC值分别为0.932、0.897、0.902、0.935和0.849,P值均小于0.01,联合这五项指标综合分析的AUC值为0.985,特异度96.87%,敏感度96.87%。在动脉期以POE+ACC+NDA组合的误判率最低(1.56%),所筛选的最佳纹理参数分别为WavEnHH_s-1(高高频小波系数转换s-1)、WavEnHH_s-2(高高频小波系数转换s-2)、135dr_ShrtREmp (短游程补偿135方向)、GrVariance (绝对梯度方差),对应的AUC值分别为0.916、0.711、0.797和0.793,P值均小于0.01,联合这四项指标综合分析的AUC值为0.916,特异度84.37%,敏感度81.25%。联合平扫+动脉期所有最佳纹理参数综合分析的AUC值为0.997,特异度93.75%,敏感度100%。结论:利用CT平扫与动脉期图像纹理分析对膀胱乳头状瘤和膀胱癌的鉴别具有一定的应用价值。

     

    Abstract: objective:To explore the value of non-enhanced CT images and enhanced CT image texture analysis in differentiating bladder papilloma from bladder cancer. Methods:64 pathologically confirmed cases of benign and malignant bladder tumors in the third Affiliated Hospital of Anhui Medical University from January 2016 to January 2020 were retrospectively included, including 32 cases of benign lesions and 32 cases of malignant lesions. All patients underwent phase III dynamic enhanced CT scan, and their non-enhanced and arterial phase images were selected for study. MaZda texture analysis software was used to extract texture features of bladder lesions on CT non-enhanced and arterial phase images. Fisher method, the minimum classification error and the minimum average correlation coefficient method(POE + ACC) and the related information measurement method(MI) were selected to identification of benign and malignant lesions of bladder 10 optimal texture characteristic value, Using Mazda bl1 tools of principal component analysis(PCA) and linear discriminant analysis(LDA) and nonlinear discriminant analysis(NDA) to select the best texture feature analysis, The minimum misjudgment rate(R) for differentiating benign from malignant bladder tumors was calculated. ROC test was conducted for the optimal texture parameters corresponding to the minimum misjudgment rate, and the quantifiable parameters with the most auxiliary differential significance were selected for differential diagnosis of benign from malignant bladder tumors. Results:Through the study found that in differentiating benign and malignant tumors of bladder images of omics, based on the non-enhanced CT images MI + NDA combination of misjudgment rate is the lowest(1.56%), The best texture parameters selected were wavenhh_s-1, Horzl_Fraction, Horzl_ShrtREmp, Sigma and Variance, with AUC values corresponding to(0.932, 0.897, 0.902, 0.935 and 0.849, all P values less than 0.01), Combined with these five indicators, the AUC value was 0.985, specificity 96.87% and sensitivity 96.87%, with statistically significant differences. At the artery stage, POE + ACC + NDA combination had the lowest misjudgment rate(1.56%). The best selected texture parameters are:wavenhh_s-1, wavenhh_S-2, 135 dr_ShrtREmp, GrVariance, AUC corresponding to(0.916, 0.711, 0.797, 0.793, all P values less than 0.01) The AUC value combined with these four indicators was 0.916, the specificity was 84.37%, and the sensitivity was 81.25%. The difference was statistically significant. The AUC value of the combined non-enhanced and enhanced best texture parameter analysis was 0.997, Specificity was 93.75%, sensitivity was 100%, the difference was statistically significant. Conclusion:The technique of texture analysis using CT non-enhanced and enhanced image has certain application value in the differentiation of bladder papilloma and bladder cancer.

     

/

返回文章
返回