Introduction

MedMiner is a platform designed to assist doctors in analyzing medical records, with a focus on patient comorbidities. Using graph-based representations, MedMiner provides a clear and intuitive visualization of the relationships between patients, diseases, and prescriptions, enabling a comprehensive view of connections between various medical conditions. This graph representation helps identify relevant disease associations, aiding doctors in exploring comorbidity patterns and potentially improving clinical management.

The project leverages Social Network Analysis (SNA) techniques to analyze health networks, applying metrics such as centrality, clustering, and other connectivity measures to identify key diseases in comorbidity networks.

Key Features

• Clinical Graph Visualization: Provides a visual representation of relationships between patients, diseases, and prescriptions, with the ability to explore the details of each node and edge in the graph.
• SNA Metrics Application: Offers essential network analysis metrics, such as betweenness, closeness, and k-core, crucial for analyzing comorbidities and identifying central or highly connected diseases.
• Interactivity and Temporal Analysis: Enables dynamic data analysis over time, allowing the observation of how comorbidities evolve for each patient.
• Custom Database Management and Loading: Allows users to dynamically upload and switch databases using structured CSV files, providing flexibility in managing different data sets.

In addition to its primary focus on comorbidity analysis, MedMiner is designed to be flexible and extendable for other medical studies. The platform includes a carousel on the welcome page, which currently showcases the ComorGraph module. This space is customizable, allowing users to add other study modules as needed, expanding the range of analyses that can be performed.

Separate AI Study

Alongside the MedMiner project, a separate artificial intelligence study has been conducted to explore the use of machine learning models aimed at predicting new relationships between patients and diseases. This study, developed separately using PyTorch Geometric, aims to enhance the platform's predictive and analytical capabilities but is not directly integrated into the main platform.

Technologies Used:

• Frontend: Developed using React, the frontend manages the user interface and allows doctors to dynamically interact with the graph. It offers a smooth and interactive experience for visualizing medical data.

• Backend: Implemented in Python, the backend communicates with a Neo4j graph database, optimized for handling graph-structured data. Through the backend, complex queries can be executed, and network analysis algorithms can be applied to extract meaningful insights from the graph.

• Database: Neo4j is used to store the relationships between patients, diseases, and prescriptions, representing them as a graph that can be queried and analyzed to uncover comorbidity patterns and other relevant information.

Project Structure

The project is organized as follows:

MedMiner/
│           # Study of machine learning models for comorbidity prediction
|── AI_Study/        # Scripts and models developed separately using PyTorch Geometric
│
├── backend/             # Backend server implementation (Python, Flask, Neo4j)
│
├── frontend/            # User interface implementation (React)
│
└── README.md            # This file

Separate Guides

Detailed instructions for dependency installation and configuration will be provided in separate README files located in the backend and frontend directories.

Authors

-Gianfranco Barba 👨‍💻

-Tullio Mansi 👨‍💻