Prediction model of tumor marker examination based on fusion of CNN and bilstm
Manfu Ma, Wenxia Wang, Hai Jia, Xia Wang, Yong Li, Sha Zhang · 2021
CNN can efficiently learn the local features of electronic medical record (EMR), ignoring the relevance between context semantics; The BiLSTM is the opposite. To solve this problem, this paper introduces attention mechanism and multilayer perceptron, combining CNN algorithm with BiLSTM algorithm. The attention mechanism is used to extract key information of electronic medical records, and the multi-layer perceptron splits local features learned by CNN, contextual semantic features learned by BiLSTM and key EMR features extracted by attention mechanism. This paper proposes a model of CNN-attention-BiLSTM for tumor marker detection prediction. It can be used in tumor marker detection and prediction. The experiment showed that compared with the model without attention mechanism and multilayer perceptron, the accuracy rate, recall rate and F1 value increased by 6.87%, 4.42% and 5.06% respectively. In this paper, this model is compared with traditional machine learning classification algorithms. Experimental results show that the accuracy of this model is 7.77% higher than that of Naive Bayes, which has the best classification effect, and 17.97% higher than that of SVM, which has the worst classification effect. Through experimental comparison, it can be concluded that the model proposed in this paper can better help doctors to assist decision-making to a certain extent.