Construction of data domain based cascade ensemble learning algorithm for pathological image diagnosis

The use of medical images is an important link in the early detection of cancer. Artificial intelligence theories and technologies such as machine learning and neural networks have gradually become important tools to assist in solving the classification problems of pathological images. However, traditional methods ignore the irreversible cost caused by misdiagnosis when the tumor is classified as other types in the prediction process, and it is difficult to obtain a classifier with high robustness for all data sources due to the differences in the manipulations of different operators in the process of slicing, staining and extracting small images. In this paper, we construct a cascade ensemble learning algorithm BDD-Boost (Based on Data Domin-Boost). A pathological image classification method of cascading multiple classifiers is proposed and used. Through the feature extraction and processing on the neurons of the preorder classifier, the adaptive threshold was generated to divide the data set. The postorder classifier was trained on the specified image set depending on the data set division result of the preorder classifier. Finally, the optimization results were obtained by integration. We verify the effectiveness of our algorithm by comparing experiments without BDD-Boost ensemble learning algorithm and with BDD-Boost ensemble learning algorithm. BDD-Boost improves the recognition performance by $1 \%$ on MobileNet_V2 and $2.2 \%$ on ResNet50. In the comparison of voting method and Bagging, it also shows better performance. The model proposed in this paper combines the advantages and characteristics of various convolutional neural networks, and has better performance and reliability than using a single convolutional neural network to directly predict the classification results of medical images under the condition of reducing computing resource consumption and calculation time. It can be used for computer-aided diagnosis and provide diagnostic reference for medical staff.