MedSynapse EHR: An Automated Summarization Platform for Electronic Health Records Based on Domain-Fine-Tuned Large Language Models

MedSynapse EHR is an open-source Flask-based full-stack application for automated summarization of electronic health records (EHR). It supports PDF/text uploads, extracts content, and generates structured clinical summaries with 8 standard chapters. Key features include role-separated user/admin consoles, optional Hugging Face Flan-T5 domain fine-tuning, and a heuristic fallback for GPU-less environments. The project aims to help medical professionals quickly access critical patient info, improving diagnostic and treatment efficiency.

Medical documents often contain large amounts of unstructured text, requiring doctors to spend significant time reading and understanding. This project addresses this pain point by automating EHR summarization, converting complex docs into structured, easy-to-read formats to boost clinical workflow efficiency.

End User Functions: Register accounts (with email uniqueness check and password hashing), manage patients (add/edit/delete), upload PDF/text EHRs (16MB limit, pypdf for PDF extraction), view extracted text and generate summaries.

Admin Functions: System metrics monitoring (user/patient/EHR counts), user management (create/edit/ban users), patient info oversight.

Structured Summary: Generates 8 chapters: Patient Info, Symptoms, Diagnosis, Drugs, Lab Reports, Treatment Plan, Doctor Notes, Follow-up Arrangements.

Tech Stack: Backend (Flask/Python), Frontend (HTML/CSS), PDF processing (pypdf), ML (Hugging Face Transformers Flan-T5), Data Storage (SQLite for dev).

Project Structure: Clear layers (app/, static/, templates/, training/, config.py, run.py etc.).

Deployment: Supports local and cloud deployment; configurable via .env file (DB connection, model path, upload limits).

The platform offers two summary modes:

1. ML Mode: Uses domain-fine-tuned Flan-T5 for high-quality summaries (requires GPU).
2. Heuristic Mode: Rule-based, no GPU needed (for resource-limited environments).

The fine-tuning enhances accuracy and medical term understanding, while the fallback ensures usability across diverse setups.

This project is a demo and should not be used for real patient care or PHI handling without proper legal review, security controls, and compliance (e.g., HIPAA). Production deployment requires: data encryption (transit/storage), access control & audit logs, patient privacy protection, regulatory compliance checks, and security vulnerability fixes.

MedSynapse EHR provides a complete, practical solution for EHR automation, covering from upload to structured output, user/admin roles, and adaptive model options. It's a valuable reference for developers/researchers exploring LLM applications in healthcare, showcasing how to integrate domain-tuned models with real-world workflows.