论文部分内容阅读
本文利用影像组学的方法预测乳腺肿瘤分子标记物雌激素受体(ER)。首先采用基于相位信息的动态轮廓模型(PBAC)对乳腺图像进行分割,其次对乳腺超声图像中肿瘤的形态、纹理、小波三个方面的404个高通量特征进行提取并予以量化,然后利用R语言以及结合最大相关最小冗余(m RMR)准则的遗传算法进行特征筛选,最后利用支持向量机(SVM)和AdaBoost进行分类判别,实现根据乳腺超声图像预测分子病理指标ER的目的。对104例临床乳腺肿瘤超声图像数据进行实验,在使用AdaBoost作为分类器的情况下得到了最优指标,即分子标记物ER的预测准确率最高可以达到75.96%,受试者操作特性曲线下的面积(AUC)最高达到79.39%。实验结果证明了利用影像组学方法预测乳腺癌ER表达情况的可行性。
In this paper, we use the method of imageology to predict the breast tumor molecular marker estrogen receptor (ER). Firstly, the dynamic contour model based on phase information (PBAC) was used to segment the breast image. Secondly, 404 high-throughput features of the tumor morphology, texture and wavelet in the breast ultrasound images were extracted and quantified. Language and genetic algorithm based on the mRMR criterion. Finally, SVM and AdaBoost are used to classify the features and predict the molecular pathology index (ER) according to the breast ultrasound images. The experimental data of 104 cases of clinical breast tumor were obtained. The optimal index was obtained when AdaBoost was used as the classifier. The prediction accuracy of molecular marker ER reached to 75.96%, and under the operating characteristic curve Area (AUC) up to 79.39%. The experimental results demonstrate the feasibility of using image analysis to predict ER expression in breast cancer.