密集卷积神经网络和辅助特征相结合的乳腺组织病理图像有丝分裂检测方法

Mitosis Detection in Breast Tissue Pathological Images by Combining Additional Features into the DenseNet

显微镜下特定大小视野范围内的平均有丝分裂个数是乳腺癌分级的一个重要指标。传统的人工检测方法耗时费力,结果受病理医生主观因素影响大,容易出错。本文提出将密集卷积神经网络(DenseNet)与辅助特征相结合,构建预测模型,以实现有丝分裂的自动检测。本文方法针对训练过程中正负样本严重不均衡问题,使用代价敏感损失函数缓解该问题。利用本文方法与其他算法对乳腺组织病理图像有丝分裂进行检测,实验结果表明,本文方法在独立测试集上的F分数为0. 801 9,高于其他方法,验证了其有效性。

Under the microscope,a key indicator of BCa grading is the average number of mitosis in a specific size field of vision.However,the traditional approach of detecting mitosis by pathologists is time-consuming and laborious.Moreover,the result is greatly influenced by subjective factors and may give a wrong diagnosis.In this work,we propose an automatic mitosis detection method by constructing the predictive model which combines additional features into the DenseNet.Furthermore,we use a costsensitive loss function to alleviate the problem of imbalance between positive and negative samples in the training process.The mitosis of breast tissue pathology was detected by the method and other algorithms.The experimental results show that the F-measure of this method is 0.801 9,which is higher than the F-measure of the other methods on the independent test set,and its effectiveness is verified.