search

yaseen28 / MedDoc-Bot

A Chat Tool for Comparative Analysis of Large Language Models in the Context of Pediatric Hypertension Guidelines
Apache License 2.0
2 stars 0 forks source link

readme

This repository contains the implementation of our full paper accepted for the IEEE EMBC 2024 Conference.

You can access the article on arXiv.

MedDoc-Bot: A Chat Tool for Comparative Analysis of Large Language Models in the Context of the Pediatric Hypertension Guideline

  1. The MedDoc-Bot interface CODE allows users to choose from four quantized Language Model Models (LLMs) to chat with multiple PDF documents.The models used for our evaluations are downloaded from huggingface (Link provided below).
  2. In our clinical use case, we assessed each model's performance by interpreting the hypertension in children and adolescents ESC guidelines PDF document. Source
  3. The original pediatric hypertension guidelines Link contain text, tables, and figures on twelve pages. We carefully transformed figures and tables into textual representations to enhance interpretation and extraction. This involves providing detailed captions, extracting numerical data, and describing visual features in text Transformed Document For Visual Element Analysis.
  4. Evaluation involved using a benchmark dataset crafted by a pediatric specialist with four years of experience in pediatric cardiology manually generated twelve questions and corresponding responses by meticulously reviewing the pediatric hypertension guidelines. Dataset.
  5. Evaluated models' accuracy, chrF, and METEOR score Detailed Results.

MedDoc-Bot Chat Tool

A Streamlit-Powered Chat Tool for interpreting Multi-PDF Document using Four Large Language Models.

Image1 Image2
Image1 Image2

MedDoc-Bot:
Manual Installation Guide Using Anaconda

1. Install Conda

https://docs.conda.io/en/latest/miniconda.html

2. Create a new conda environment

conda create -n MedDoc-Bot python=3.11
conda activate MedDoc-Bot

3. Install Pytorch

System GPU Command
Windows NVIDIA pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
Windows CPU only pip3 install torch torchvision torchaudio

The up-to-date commands can be found here: https://pytorch.org/get-started/locally/.

For NVIDIA, you also need to install the CUDA runtime libraries:

conda install -y -c "nvidia/label/cuda-12.1.1" cuda-runtime

4. Install the web UI

git clone https://github.com/yaseen28/MedDoc-Bot
cd MedDoc-Bot
pip install -r requirements.txt

5. Download the Four Pre-Quantised Language Models to the Project Folder

(i) Llama-2 {Version: llama-2-13b.Q5_K_S.gguf} Link
(ii) MedAlpaca {Version: medalpaca-13b.Q5_K_S.gguf} Link
(iii) Meditron {Version: meditron-7b.Q5_K_S.gguf} Link
(iv) Mistral {Version: mistral-7b-instruct-v0.2.Q5_K_M.gguf} Link

NOTE!! Please ensure that you rename the model file to match the name listed in the 'Select Model' dropdown in the browser.

6. Start the MedDoc-Bot

conda activate MedDoc-Bot
cd MedDoc-Bot
streamlit run Main_MedDoc-Bot.py

You can now view your Streamlit app in your browser.

Local URL: http://localhost:8501

7. Provide Default Username and Password

User
User@123

If you find our work useful, please consider citing it.

@misc{jabarulla2024meddocbot, title={MedDoc-Bot: A Chat Tool for Comparative Analysis of Large Language Models in the Context of the Pediatric Hypertension Guideline}, author={Mohamed Yaseen Jabarulla and Steffen Oeltze-Jafra and Philipp Beerbaum and Theodor Uden}, year={2024}, eprint={2405.03359}, archivePrefix={arXiv}, primaryClass={cs.CL} }