经直肠前列腺超声图像纹理分析鉴别前列腺良恶性疾病的价值探讨

Value of the texture analysis of transrectal prostate contrast-enhanced ultrasound images in differentiating benign and malignant prostate diseases

目的探讨经直肠前列腺超声图像纹理分析鉴别前列腺良、恶性疾病的临床价值。方法回顾性分析经病理证实的84例前列腺癌患者和41例前列腺良性病变患者的超声图像,对照经直肠前列腺超声造影发现的外腺病灶,在灰阶超声及超声造影图像上手动勾画病灶感兴趣区域,提取纹理特征,然后进行特征筛选,筛选后的特征进行组间差异性分析并绘制ROC曲线,选取超声造影及灰阶超声最佳纹理特征联合诊断,比较诊断效能有无统计学差异。结果筛选后的特征主要来源于灰度共生矩阵(Gray-level Co-occurrence Matrix,GLCM),AUC>0.75的特征均来自于GLCM。灰阶超声最佳纹理特征、超声造影最佳纹理特征、最佳纹理特征联合诊断的AUC、敏感度、特异度分别为0.773、0.817、0.899,65.48%、91.67%、88.10%,85.37%、60.98%、82.93%。最佳纹理特征单独诊断前列腺癌的AUC差异无统计学意义(P>0.05),联合诊断效能均高于单独诊断,差异有统计学意义(P<0.05)。结论基于灰度共生矩阵的经直肠前列腺超声图像纹理分析对前列腺癌有较好的诊断效能,可作为辅助临床诊断前列腺癌的工具,具有潜在临床应用价值。

Objective To investigate the clinical value of texture analysis of transrectal prostate contrast-enhanced ultrasound images in differentiating benign and malignant prostate diseases.Methods This study retrospectively analyzed the contrast-enhanced ultrasound images of 84 patients with pathologically confirmed prostate cancer and 41 patients with benign prostate disease.According to the external gland lesions found by transrectal prostate contrast-enhanced ultrasound,the concerned regions of the lesions were manually delineated on the gray-scale ultrasound and contrast-enhanced ultrasound images.Texture features were extracted,and then the features were screened.After screening,the differences in texture features between groups were analyzed and the ROC curves were drawn.The best texture features of contrast-enhanced ultrasound and gray-scale ultrasound images were selected to make a combined diagnosis.The diagnostic efficacy was compared to observe whether there was statistical difference.Results The features screened were mainly derived from gray-level Co-occurrence Matrix(GLCM),and all the features with AUC>0.75 were derived from GLCM.The AUC,sensitivity and specificity of the best texture features of the grey-scale ultrasonic image,the contrast-enhanced ultrasound images and the combined texture features were 0.773,0.817,0.899,65.48%,91.67%,88.10%,85.37%,60.98%,82.93%,respectively.There was no significant difference in the AUC of the best texture feature in the diagnosis of prostate cancer alone(P>0.05),but the combined diagnosis had higher efficiency than the single diagnosis,and the difference was statistically significant(P<0.05).Conclusion Texture analysis of transrectal prostate ultrasound images based on GLCM has good diagnostic efficacy for prostate cancer,which can be used as an auxiliary tool for clinical diagnosis of prostate cancer.And it has potential value of clinical application.