基于联合训练的分类器的乳腺癌图像分类

Breast cancer image classification based on joint training classifier

摘要:

利用机器学习的乳腺癌组织病理图像诊断节省了大量的人力物力，因此提高乳腺癌组织病理图像识别准确率有很好的现实意义。针对单一分类器和集成学习分类器模型观测域有限容易陷入局部最优的问题，提出一种基于联合训练的分类器模型。通过单一分类器相互影响扩大观测感知域来寻找损失最小的估计点，根据估计点来迭代优化超参数进而联合训练出拟合性能最好的分类器，这样既汲取不同分类器模型的可取之处来增强泛化能力，又加大了模型观测域在可以更快的得到全局最优的同时提升了识别准确率。实验表明，提出的联合训练的分类器能够提升乳腺癌组织病理学图像的分类性能，在不同放大倍数40×、100×、200×、400×下图像良恶性分类准确率分别为99.67%、98.08%、99.01%、96.34%。

Abstract:

The histopathological image diagnosis of breast cancer using machine learning saves a lot of manpower and material resources, so it is of good practical significance to improve the accuracy of histopathological image recognition of breast cancer. Aiming at the problem that the observation domain of single classifier and ensemble learning classifier model is limited and easy to fall into local optimality, a classifier model based on joint training is proposed. Through the mutual influence of a single classifier, the observation perception domain is expanded to find the estimated point with the least loss, and the hyperparameters are iteratively optimized according to the estimated point and then jointly trained to jointly train the classifier with the best fitting performance, which not only absorbs the desirability of different classifier models to enhance the generalization ability, but also increases the model observation domain to obtain the global optimal faster while improving the recognition accuracy. Experiments show that the proposed jointly trained classifier can improve the classification performance of breast cancer histopathological images,and the accuracy of benign and malignant classification of images at different magnifications of 40×, 100×, 200× and 400× is 99.67%, 98.08%, 99.01% and 96.34%, respectively.