부산대학교 도서관

The International Classification of Diseases (ICD) is a vital tool used in clinical and health management, providing codes for disease classification. The use of deep learning techniques to automatically extract valuable information from medical records and assist in coding has gained significant attention due to the increasing volume of medical data and the advancement of precision medicine research. However, current coding methods face challenges such as the large candidate space for disease coding and imbalanced code distribution. This study focuses on these challenges and proposes a hierarchical ICD automatic coding method. By introducing a Transformer-based hierarchical path propagation mechanism, the study effectively captures the relationships between disease codes at different hierarchical levels and reduces the candidate space for coding. Experimental results demonstrate the method’s efficacy in information extraction and coding improvement.