Report Q L base

Introduction

This repository provides the code for the study "Application of Deep Learning in Generating Structured Radiology Reports: A Transformer-Based Technique." The goal is to convert free-text radiology reports into structured data using natural language processing (NLP) and deep learning (DL) techniques. The project employs a transformer-based architecture for fine-grained named entity recognition (NER) to extract clinical information from radiology reports. The study demonstrated improved performance over previous models such as artificial neural networks (ANN) and convolutional neural networks (CNN), using metrics like ROUGE and BLEU scores.

Architecture

The model utilizes the Text-to-Text Transfer Transformer (T5) and its domain-specific adaptation, Scifive, for clinical information extraction. The architecture processes free-text reports and transforms them into structured formats. The dataset includes 88 annotated abdominopelvic sonography reports, structured according to a specific information schema.

Training

To fine-tune the language model, the project setup involves installing necessary dependencies in a virtual environment. The fine-tuning process is executed with a script provided in the repository, which adjusts the pre-trained T5 model to the task of extracting and structuring clinical data.

Guide: Running Locally

1. Environment Setup:

   • Create a virtual environment: virtualenv .venv
   • Activate the environment: source .venv/bin/activate

2. Install Dependencies:

   • Run pip install -r requirements.txt to install all required packages.

3. Fine-Tuning:

   • Execute python script/fit.py to start fine-tuning the model.

4. Testing:

   • Run python script/test.py to evaluate the model on the test dataset.

5. Inference:

   • Use the provided Jupyter notebook for inference: notebooks/predict_reportql.ipynb.

For optimal performance, consider using cloud GPUs from providers such as AWS, Google Cloud, or Azure.

License

The project is licensed under the MIT License. For more details, refer to the LICENSE file in the repository.