基于机器学习对乳腺癌超声图像数据分类

Classification of breast cancer ultrasound image data based on machine learning

目的探讨机器学习(Machine Learning)在乳腺癌超声图像数据分类中的应用价值。方法使用Online Medical Images数据库中的超声数据,整理出727张超声影像数据,其中502张乳腺癌超声数据,225张健康人乳腺超声数据,将其数据维度统一设置为64×64大小。按照类别将60%的数据作为训练集数据,40%的数据作为测试集数据。通过支持向量机(Support Vector Machine,SVM)方法,选择不同特征向量对训练集进行模型训练,在独立验证集数据中验证模型性能,包括曲线下面积(AUC)以及相应的灵敏度和特异性。结果共找到291个特征向量,将其作为SVM的分类支持向量,将其中的197个特征支持向量作为肿瘤的特征支持向量,将94个特征支持向量作为正常的支持向量。对测试集数据进行分析,曲线下面积(AUC)为0.88,相应的灵敏度为90.31%,特异性为87.25%。结论机器学习在乳腺癌超声图像数据分类中具有重要应用价值,具有较高灵敏度和特异性,适合推广应用。

Objective To explore the application value of machine learning(Machine Learning)in breast cancer ultrasound image data classification.Methods Use the ultrasound data in the Online Medical Images database to sort out 727 ultrasound image data,including 502 breast cancer ultrasound data and 225 healthy breast ultrasound data,and organize them into a data dimension of 64×64.According to the category,60%of the data is used as the training set data,and 40%of the data is used as the test set data.Through the Support Vector Machine(SVM)method,different feature vectors are selected for model training on the training set,and the performance of the model is verified in the independent validation set data,including the area under the curve(AUC)and the corresponding sensitivity and specificity.Results A total of 291 feature vectors were found,which were used as the classification support vectors of SVM,197 feature support vectors were used as tumor feature support vectors,and 94 feature support vectors were used as normal support vectors.Analyzing the test set data,the area under the curve(AUC)is 0.88,the corresponding sensitivity is 90.31%,and the specificity is 87.25%.Conclusion Machine learning has important application value in breast cancer ultrasound image data classification,with high sensitivity and specificity,and is suitable for popularization and application.