Interpretable ICD Code Classification with Faithful Sentence Extraction

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Health & Wellbeing — Health & Medical Research, Medical Devices & Health Technology, Clinical Care & Medical Practice · Depth: Advanced, short

Summary

A new method addresses the lack of faithful explanations in Transformer-based models like PLM-CA for automatic ICD code classification in electronic medical records. Researchers jointly train a sentence extractor and an ICD code classifier, ensuring predictions rely solely on the extracted sentences. This approach provides faithful rationales for each predicted code, significantly reducing the effort needed to inspect lengthy medical records. Experiments on the MIMIC-III dataset demonstrate that this method achieves performance comparable to a full-record transformer baseline while processing only a small fraction of the document, offering a more interpretable solution for clinical applications.

Key takeaway

For AI Scientists developing clinical NLP systems, traditional attention weights often fail to provide faithful explanations for critical predictions. You should consider integrating a jointly trained sentence extractor and classifier to ensure predictions are directly tied to explicit textual evidence. This approach enhances model interpretability and trustworthiness, which is crucial for reducing manual verification efforts and increasing adoption in sensitive medical record applications.

Key insights

Faithful sentence extraction provides interpretable rationales for ICD code predictions, improving trust in AI.

Principles

• Faithful explanations are critical for medical AI adoption.
• Joint training can directly link predictions to evidence.
• Reduced input can maintain performance with interpretability.

Method

Jointly train a sentence extractor and an ICD code classifier. Ensure predictions are based exclusively on the extracted sentences to provide faithful rationales.

In practice

• Use extracted sentences as direct evidence for code assignments.
• Reduce manual review time for medical record coding.
• Implement joint training for explainable NLP tasks.

Topics

• ICD Code Classification
• Interpretable AI
• Sentence Extraction
• Electronic Medical Records
• Transformer Models
• MIMIC-III

Best for: NLP Engineer, AI Scientist, Research Scientist, AI Ethicist

Related on AIssential

• Interpretable Discriminative Text Representations via Agreement and Label Disentanglement — Interpretable text representations require features with high annotator agreement and clear disentanglement from prediction targets for practical auditability.
• Quantifying Explanation Consistency: The C-Score Metric for CAM-Based Explainability in Medical Image Classification — The C-Score quantifies explanation consistency in CAM methods, revealing model instability beyond classification metrics.
• ChatHealthAI: Aligning Electronic Health Record Representations with Large Language Models for Grounded Clinical Reasoning — ChatHealthAI integrates EHR representations with LLMs for grounded, interpretable clinical reasoning and accurate patient prediction.
• SenFlow: Inter-Sentence Flow Modeling for AI-Generated Text Detection in Hybrid Documents — SenFlow improves AI-generated text detection in hybrid documents by modeling inter-sentence dependencies and using a new benchmark.
• Natural Language Autoencoders: Turning Claude’s thoughts into text — Natural Language Autoencoders translate AI model internal activations into human-readable text, revealing hidden thoughts and motivations.
• Application of integrated gradients explainability to sociopsychological semantic markers — Integrated gradients reveal word-level contributions to sociopsychological text classification, even with limited data.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.