license: mit
language:
- en
base_model:
- meta-llama/Llama-3.1-8B-Instruct
pipeline_tag: text-classification
tags:
- medical
RadReportX
Model description
Llama3.1-8B-instruct model fine tuned on synthetic data. There are two tasks that this model can achieve. The first task is an open-ended question, which is to detect phrases in a radiology report that represents an ICD-10 code. There is no restriction about the underlying disease. The second task is to detect disease out of 13 candidates from a radiology report. The candidate diseases are [Atelectasis, Cardiomegaly, Consolidation, Edema, Enlarged Cardiomediastinum, Fracture, Lung Lesion, Lung Opacity, Pleural Effusion, Pleural Other, Pneumonia, Pneumothorax, Support Devices]. When there are no diseases out of the candidates, the model will output 'Normal'.
Training set and training process
There are two sources of training data. The first set is generated by GPT4o. The second source comes from MIMIC-CXR dataset (https://arxiv.org/pdf/1901.07042), with labels being extracted by Negbio algorithm. The training is conducted using torchtune framework (https://github.com/pytorch/torchtune). For details, please refer to our paper listed below.
How to use
Please refer to https://github.com/bionlplab/RadReportX
Paper
https://arxiv.org/pdf/2409.16563
Citation
@article{wei2024enhancing, title={Enhancing disease detection in radiology reports through fine-tuning lightweight LLM on weak labels}, author={Wei, Yishu and Wang, Xindi and Ong, Hanley and Zhou, Yiliang and Flanders, Adam and Shih, George and Peng, Yifan}, journal={arXiv preprint arXiv:2409.16563}, year={2024} }
Disclaimer
This tool shows the results of research conducted in the Computational Biology Branch, NCBI. The information produced on this website is not intended for direct diagnostic use or medical decision-making without review and oversight by a clinical professional. Individuals should not change their health behavior solely on the basis of information produced on this website. NIH does not independently verify the validity or utility of the information produced by this tool. If you have questions about the information produced on this website, please see a health care professional. More information about NCBI's disclaimer policy is available.