基于随机森林的乳腺肿瘤诊断研究

Research on Diagnosis of Breast Cancer Based on Random Forest

采用人工智能方法对乳腺肿瘤进行自动诊断实质是对乳腺显微图像进行模式分类识别问题,比传统的人工诊断方法具有更高的准确率和效率,从而提高肿瘤治疗效果。基于随机森林的分类器具有良好的泛化性能,首先讨论了随机森林模型的建立,然后利用训练好的模型对乳腺肿瘤数据进行分类测试,最后讨论了影响随机森林分类器性能的因素以及如何选择随机森林里的决策树的数量。仿真实验表明,利用随机森林分类器对乳腺肿瘤进行分类识别比采用BP、LVQ神经网络、决策树方法可以获得更好的泛化性能。

Using artificial intelligence methods for automatic diagnosis is essentially a pattern recognition issue for the microscope images of the breast tumor, compared with the traditional diagnostic methods ,which have a higher accuracy and efficiency, so as to improve the therapeutic effects. Because the random forest classifier has better generalization performance, the random forest model establishing process is first discussed, and then the breast canc- er test data set is classified by the trained model. Finally,the factors affecting the random forest performance as well as how to choose the number of the decision trees are analyzed. The simulation result is showed that this algorithm can obtain better generalization performance than using BP,LVQ neural network, the decision tree methods for breast cancer diagnosis.