基于影像组学预测乳腺癌雌激素受体表达情况的可行性分析

发布时间：2018-05-29 08:35

本文选题：影像组学 + 高通量特征　；参考：《生物医学工程学杂志》2017年04期

【摘要】：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。
[Abstract]:In this paper, estrogen receptor ERA, a molecular marker of breast tumors, was predicted by imaging method. First, the dynamic contour model (PBAC) based on phase information is used to segment the breast image. Secondly, 404 high-throughput features of breast ultrasound image are extracted and quantified in three aspects: shape, texture and wavelet. Then we use R language and genetic algorithm combined with maximum correlation minimum redundancy (MRR) criterion for feature screening. Finally, support vector machine (SVM) and AdaBoost are used to classify and discriminate, and the purpose of predicting molecular pathological index ER based on breast ultrasound image is realized. The ultrasound image data of 104 cases of clinical breast tumors were tested, and the optimal index was obtained by using AdaBoost as the classifier. That is, the prediction accuracy of molecular marker ER can reach 75.96, and the area under the operating characteristic curve of subjects can reach 79.39. The experimental results proved the feasibility of predicting ER expression in breast cancer by imaging.
【作者单位】：复旦大学电子工程系;上海市医学图像处理与计算机辅助手术重点实验室;复旦大学附属肿瘤医院超声诊断科;
【基金】：国家重点基础研究发展计划(2015CB755500)
【分类号】：R737.9;TP391.41

【相似文献】