Skip to content

πŸ“„ Preprint submitted to Artificial Intelligence in Medicine πŸ₯ - "MediCARE: Medical Collaborative Agents REasoning over Interpretable Heterogeneous Graphs" πŸ’ŠπŸ“Š - University of Naples "Federico II" πŸŽ“

Notifications You must be signed in to change notification settings

LaErre9/MediCARE

Β 
Β 

Repository files navigation

πŸ’Š MediCARE Framework

MediCARE is an advanced framework designed to revolutionize personalized medical recommendations through the integration of Machine Learning techniques and Graph Neural Networks (GNNs). Developed during MSc Thesis Internship at University of Naples "Federico II" in collaboration with PicusLab, this framework aims to optimize clinical decision-making by leveraging Electronic Health Records (EHRs) data, within the MIMIC-III database (noteevents). The overall framework has been integrated within the open-source PyHealth library (check the fork repo here).

✨ Features

  • Personalized Recommendations: MediCARE utilizes state-of-the-art machine learning algorithms to generate personalized medical recommendations tailored to individual patient profiles;
  • Graph Representation: The framework employs Heterogeneous Graph representations to capture complex relationships between medical concepts extracted from EHR data through Named Entity Recognition and Entity Linking (NER+EL), ensuring a comprehensive understanding of patient health;
  • Integration of External Knowledge: MediCARE enriches patient graphs with static Medical Knowledge Graphs, such as DRKG and SympGAN, enhancing the accuracy and depth of medical recommendations by introducing additional medical concepts and relationships;
  • Explainable AI (XAI): The framework incorporates eXplainable Artificial Intelligence (XAI) methods such as GNNExplainer and Integrated Gradients to interpret the recommendations and internal mechanims of the GNN models;
  • Multi Agent Collaboration LLMs: MediCARE with this solution generate, discuss, and revise comprehensive assessments to explain the predictions of the model.

βš™οΈ Project Workflow

MediCARE

πŸ› οΈ Tools used

PyG PyHealth MedCat Python OpenAI AutoGen Streamlit

⚠️ Warning

It is essential to note that MIMIC-III contains sensitive patient data and, therefore, must be handled with the utmost care and in compliance with privacy regulations and institutional policies. Before using the dataset, carefully review and adhere to the guidelines provided by the MIMIC team and consult your local ethics committee.

βœ… Project realised for demonstration and educational purposes only

Copyright Β© 2024 - MediCARE Framework project carried out for the MSc Internship held at the University of Naples "Federico II". Realised for demonstration and teaching purposes only.
Antonio Romano, Giuseppe Riccio

About

πŸ“„ Preprint submitted to Artificial Intelligence in Medicine πŸ₯ - "MediCARE: Medical Collaborative Agents REasoning over Interpretable Heterogeneous Graphs" πŸ’ŠπŸ“Š - University of Naples "Federico II" πŸŽ“

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Jupyter Notebook 98.1%
  • Python 1.9%