Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TraceCoder: Towards Traceable ICD Coding via Multi-Source Knowledge Integration

Mucheng Ren, He Chen, Yuchen Yan, Danqing Hu, Jun Xu, Xian Zeng

TL;DR

TraceCoder tackles automated ICD coding by addressing semantic gaps between clinical text and codes, interpretability, and long-tail code distribution. It proposes a dynamic, multi-source knowledge integration framework that combines UMLS, Wikipedia, and LLM-derived content with a hybrid attention mechanism to ground predictions in external evidence. Key contributions include a dynamic knowledge matching module, three attention streams (LSA, LCCA, KCCA), and a CNN-based multi-label predictor, with ablation and case studies validating effectiveness. On MIMIC-III and MIMIC-IV, TraceCoder achieves state-of-the-art results while offering improved traceability and explainability, supporting reliable real-world deployment.

Abstract

Automated International Classification of Diseases (ICD) coding assigns standardized diagnosis and procedure codes to clinical records, playing a critical role in healthcare systems. However, existing methods face challenges such as semantic gaps between clinical text and ICD codes, poor performance on rare and long-tail codes, and limited interpretability. To address these issues, we propose TraceCoder, a novel framework integrating multi-source external knowledge to enhance traceability and explainability in ICD coding. TraceCoder dynamically incorporates diverse knowledge sources, including UMLS, Wikipedia, and large language models (LLMs), to enrich code representations, bridge semantic gaps, and handle rare and ambiguous codes. It also introduces a hybrid attention mechanism to model interactions among labels, clinical context, and knowledge, improving long-tail code recognition and making predictions interpretable by grounding them in external evidence. Experiments on MIMIC-III-ICD9, MIMIC-IV-ICD9, and MIMIC-IV-ICD10 datasets demonstrate that TraceCoder achieves state-of-the-art performance, with ablation studies validating the effectiveness of its components. TraceCoder offers a scalable and robust solution for automated ICD coding, aligning with clinical needs for accuracy, interpretability, and reliability.

TraceCoder: Towards Traceable ICD Coding via Multi-Source Knowledge Integration

TL;DR

TraceCoder tackles automated ICD coding by addressing semantic gaps between clinical text and codes, interpretability, and long-tail code distribution. It proposes a dynamic, multi-source knowledge integration framework that combines UMLS, Wikipedia, and LLM-derived content with a hybrid attention mechanism to ground predictions in external evidence. Key contributions include a dynamic knowledge matching module, three attention streams (LSA, LCCA, KCCA), and a CNN-based multi-label predictor, with ablation and case studies validating effectiveness. On MIMIC-III and MIMIC-IV, TraceCoder achieves state-of-the-art results while offering improved traceability and explainability, supporting reliable real-world deployment.

Abstract

Automated International Classification of Diseases (ICD) coding assigns standardized diagnosis and procedure codes to clinical records, playing a critical role in healthcare systems. However, existing methods face challenges such as semantic gaps between clinical text and ICD codes, poor performance on rare and long-tail codes, and limited interpretability. To address these issues, we propose TraceCoder, a novel framework integrating multi-source external knowledge to enhance traceability and explainability in ICD coding. TraceCoder dynamically incorporates diverse knowledge sources, including UMLS, Wikipedia, and large language models (LLMs), to enrich code representations, bridge semantic gaps, and handle rare and ambiguous codes. It also introduces a hybrid attention mechanism to model interactions among labels, clinical context, and knowledge, improving long-tail code recognition and making predictions interpretable by grounding them in external evidence. Experiments on MIMIC-III-ICD9, MIMIC-IV-ICD9, and MIMIC-IV-ICD10 datasets demonstrate that TraceCoder achieves state-of-the-art performance, with ablation studies validating the effectiveness of its components. TraceCoder offers a scalable and robust solution for automated ICD coding, aligning with clinical needs for accuracy, interpretability, and reliability.
Paper Structure (24 sections, 11 equations, 4 figures, 3 tables)

This paper contains 24 sections, 11 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Attention Mechanisms for ICD Coding
  4. External Knowledge Integration
  5. Transformer-Based Architectures
  6. Method
  7. Context Encoding
  8. Dynamic Multi-source Knowledge Matching
  9. Code Synonyms from UMLS
  10. Wikipedia Knowledge
  11. Insights from LLMs
  12. Hybrid Attention Integration
  13. Label-wise Self-Attention (LSA)
  14. Label-Context Cross-Attention (LCCA)
  15. Multi-Label Prediction
  16. ...and 9 more sections

Figures (4)

  • Figure 1: Illustration of Auto-ICD coding task.
  • Figure 2: The flowchart of proposed TraceCoder framework.
  • Figure 3: Ablation study results on the impact of different attention mechanisms in the MIMIC-III-50 test set.
  • Figure 4: Case study illustrating TraceCoder's traceable evidence-grounded prediction.