Enhancing ICD classification with semantic embedding rectification and long-tail refinement

Yuhao Wu

doi:10.1016/j.knosys.2025.114530

Enhancing ICD classification with semantic embedding rectification and long-tail refinement

2025-09-29

0

PRE

AI

DOI：10.1016/j.knosys.2025.114530

Original

Original request for help

Save

Abstract

En 中文

• We propose RoSimTail-ICD, an end-to-end framework for addressing semantics and label imbalance in ICD coding. • We design the SSDC module to align predicted semantics with standard ICD definitions. • We propose the MATR module to enhance rare code learning via adaptive tail refinement. • RoSimTail-ICD achieves the best performance on five MIMIC datasets compared to SOTA methods.

Enhancing ICD classification with semantic embedding rectification and long-tail refinement

Abstract

Journal

Organization