利用超声图像纹理特征鉴别甲状腺实性结节良恶性的研究

The use of thyroid ultrasound imageing texture features in diagnosing the nature of thyroid solid nodules

目的利用灰度共生矩阵法提取B超图像上甲状腺实性结节病灶区的纹理特征值,并建立Logistic预测模型,并探讨该模型在鉴别甲状腺实性结节良恶性中的可行性。方法收集经手术证实的甲状腺实性结节患者临床资料,从其超声图像中提取结节区域灰度共生矩阵纹理特征值,并将该特征值作为自变量,结节的良恶性作为因变量拟合Logistic预测模型。利用10折交叉验证对预测模型进行效果评价,并绘制ROC曲线。结果 Logistic回归模型对甲状腺实性结节良恶性预测的准确率为82%,ROC曲线下面积(AUC)为0.89。结论利用甲状腺实性结节病灶超声图像灰度共生纹理特征值建立的二分类Logistic回归模型能够对甲状腺实性结节的良恶性做出较准确的判断。

Objective To evaluate the value of using texture‐based gray‐level co‐occurrence matrix （GLCM ） features ex‐tracted from thyroid ultrasound images to build logistic model for differentiating the nature of thyroid nodules .Methods We collected 94 cases patients who suffered from the thyroid nodules and accepted thyroidectomy .GLCM was used to ex‐tract texture features from their ultrasound images .Then ,we used the features as independent variables and the nature of nodules as dependent variables to build logistic model .10‐fold cross‐validation was used to evaluate the performance of the model and drew the ROC curve .Results The accuracy of the logistic regression was 82% ,and the area under the ROC curve（AUC）was 0 .89 .Conclusion The binary logistic regression built with GLCM features extracted from solid thyroid ultrasound images is useful in diagnosing the nature of thyroid solid nodules .