← Back to all news
arXiv AI By Stefan Lenz, Lakisha Ortiz Rosario, Georg Vollmar, Arsenij Ustjanzew, Fatma Alickovic, Thomas Kindler, Torsten Panholzer

Instruction-Tuning Large Language Models with Medical Classification Data Improves ICD Coding of German Tumor Diagnoses

Read the original on arXiv AI →

The study demonstrates that instruction‑tuning open‑weight large language models (LLMs) with over 500,000 public question‑answer pairs derived from ICD‑10‑GM, ICD‑O‑3, and OPS catalogues markedly improves ICD coding accuracy for German tumor diagnoses. After fine‑tuning, ICD‑10‑GM exact accuracy rose from 1.4–24 % to 41–58 %, and partial accuracy from 31–74 % to 73–83 %. The approach also eliminated malformed code outputs and achieved 99 % tumor‑diagnosis recognition, with performance scaling positively with model size but narrowing the gap between small and large models.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv AI.

Related stories

arXiv Computation and Language
Oct 22

Is a 9B Decision Foundation Model Needed for Symptom-to-Disease Classification? Fine-Tuning Clef-Flash Against TF-IDF Baselines

The paper evaluates whether a 9‑billion parameter decision foundation model, Clef‑Flash, is necessary for symptom‑to‑disease classification. Fine‑tuning the model on the 24‑class Symptom2Disease benchmark yields 95.42% accuracy, but simple TF‑IDF linear classifiers achieve higher accuracy (98.75%) with far faster inference. Ablation studies show that predictions can hinge on single descriptive clauses, indicating that the benchmark may not fully capture the model’s capabilities.

By S. M. Mahbub Uddin Ahmed
llmsfine-tuningefficiencybenchmarks
More like this →