Disease Phenotype Classification Model Based on Multi-channel Deep Supervised Bi-LSTM

Phenotypic classification of diseases refers to the process of categorizing diseases into different classes based on their manifestations or characteristics, aiding doctors in diagnosing diseases and developing treatment plans. Current classification methods are challenged by issues such as strong subjectivity, high instability, and low data utilization. To conduct phenotypic classification of diseases more effectively, this paper proposed a disease phenotypic classification model based on multi-channel deep supervised Bi-LSTM. The proposed model integrates structured and unstructured data for classification, uses the multi-channel module to integrate temporal information from multi-variable time sequence data, and utilizes Bi-LSTM module to capture long-term dependencies in single-channel sequence data. Additionally, a deep supervision module is used to accelerate model convergence and improve generalization capability. The experimental results on the MIMIC-III dataset indicate that the proposed model achieves higher classification accuracy compared to various baseline models, with a smaller parameter size and shorter training time, thereby demonstrating the effectiveness of the method presented in this paper.