MedSyn: LLM-based Synthetic Medical Text Generation Framework

The repository contains modules for generating clinical notes in Russian based on the target ICD-10 code.
The proposed framework utilizes symptoms sampled from a medical knowledge graph and real clinical note examples for a target ICD code.
Generating clinical notes without prior data is possible, but results may differ from those presented in the paper.

Data:

We provide synthetic dataset and instruction fine-tuning dataset via HF datasets:

• Synthetic dataset
• IFT dataset

Examples:

GPT-4 generation example:

LLaMA-7b generation example:

Fine-tuned LLaMA-7b:

We provide a fine-tuned LLaMA-7b model checkpoint; it can be downloaded via the link.

Citation:

@InProceedings{10.1007/978-3-031-70381-2_14,
author="Kumichev, Gleb
and Blinov, Pavel
and Kuzkina, Yulia
and Goncharov, Vasily
and Zubkova, Galina
and Zenovkin, Nikolai
and Goncharov, Aleksei
and Savchenko, Andrey",
editor="Bifet, Albert
and Krilavi{\v{c}}ius, Tomas
and Miliou, Ioanna
and Nowaczyk, Slawomir",
title="MedSyn: LLM-Based Synthetic Medical Text Generation Framework",
booktitle="Machine Learning and Knowledge Discovery in Databases. Applied Data Science Track",
year="2024",
publisher="Springer Nature Switzerland",
address="Cham",
pages="215--230",
isbn="978-3-031-70381-2"
}