FirsthandHealth
medRxiv PreprintsInternational19 November 2026

Fine-tuned encoders, generative large language models and typed decisions for classifying findings in chest CT reports: paired English-Japanese internal testing and external testing on naturally written Japanese reports

This is an official announcement record

Firsthand records what medRxiv Preprints announced and links to the original. The wording below is theirs, not ours.

Purpose: To compare fine-tuned encoders, generative large language models (LLMs) and typed decisions (per-finding yes/no questions answered with a probability) for classifying 18 findings in chest CT reports, and to identify when typed decisions help. Methods: This retrospective study used public data. The internal test set comprised 150 CT-RATE reports with paired Japanese translations at three quality levels; the external test set, 243 naturally written Japanese reports with human annotations (RR-Findings). We compared ModernBERT encoders fine-tuned on 20,444 reports, generative LLMs (gpt-6-
Source: medRxiv Preprints
Read the official announcement

Opens www.medrxiv.org

More from medRxiv Preprints

This content is for informational purposes only and is not medical advice. It is not intended to diagnose, treat, cure, or prevent any disease. Consult a healthcare professional before starting any supplement, treatment, or program: especially if you are pregnant, nursing, taking medication, or managing a health condition.