LA-TextCNN-BiLSTM: A classification model for ICD-11

A model for ICD-11 automatic coding is proposed for the problem of ICD-11 automatic coding to further improve the coding performance. This study proposes an ICD-11 automatic encoding model based on label attention mechanism to address the issues of long-distance dependency and Chinese semantic diversity. This model utilizes MC-BERT to enhance the semantic representation of electronic medical records and further supplements the semantic information of the model through label attention mechanism, reducing the interference of redundant information in the input text on the model. Finally, the deep neural network framework is used to improve the semantic extraction ability of the model and complete the final classification. The experimental dataset is constructed based on real clinical electronic medical records, and the model empirical and comparative experiments are carried out, with an accuracy rate of 83.86 %, a macro-averaged F1 value of 75.82 %, and a micro-averaged F1 of 82.83 %. It is better than the classification methods proposed by previous authors. The ICD-11 automatic coding model based on labeled attention mechanism proposed in this paper can effectively improve the performance of automatic coding of medical records. • The study combines the labeling information of ICD-11 with an automatic text classification model to improve the classification performance. • The first exploration of ICD-11 classification in the field of thyroid diseases, and the classification accuracy reached 83.56 %. • The present work further enhances the semantic representation of the Chinese text by MC-BERT.