Researchers have developed SemICD, a novel framework designed to improve the accuracy of generative models in medical coding. Traditional methods using LLMs to predict International Statistical Classification of Diseases (ICD) codes face challenges due to the hierarchical structure of ICD codes not always aligning with clinical relationships. SemICD addresses this by training LLMs to generate Semantic IDs (SIDs), which are hierarchically clustered embeddings that allow semantically related codes to share prefixes. Experiments on MIMIC-III and MIMIC-IV datasets demonstrated that SemICD significantly outperforms direct code generation, showing improvements of 1.34-8.00 Micro-F1 points across four different LLMs. AI
IMPACT This research offers a method to improve the precision of LLMs in specialized domains like medical coding, potentially leading to more accurate clinical documentation and data analysis.
RANK_REASON The cluster describes a new framework and experimental results published in a paper, fitting the research bucket. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Hugging Face Daily Papers →
- International Statistical Classification of Diseases and Related Health Problems
- MIMIC-III
- MIMIC-IV
- SemICD
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
