- Introduction
- Key Features
- Models Used
- System Architecture
- Installation
- Usage
- Evaluation & Performance
- Fine-tuned Models
- Future Enhancements
- License
The rapid increase in polypharmacy highlights the need for intelligent tools to safely manage multiple medications, especially for chronic patients who face an increased risk of drug interactions. Patients often receive prescriptions from different doctors, leading to confusion, non-compliance, and potential medication errors. The complexities of drug interactions, varied medication schedules, and lack of knowledge about prescription components contribute to these challenges.
Traditionally, healthcare technology focused on digitizing medical records and improving diagnostic accuracy. Electronic Health Records (EHRs) revolutionized patient data management, while advancements in diagnostic imaging, such as MRI and CT scans, enhanced disease detection. However, many existing healthcare platforms provide only general information and fail to offer personalized, real-time advice, particularly for managing complex medication regimens.
In recent years, Artificial Intelligence (AI) and Machine Learning (ML) have transformed healthcare. AI-powered platforms like Med-PaLM and ChatGPT APIs have improved patient query responses and medical knowledge distribution. However, these systems often lack personalization, real-time adaptability, and multimodal input capabilities, limiting their effectiveness in intricate patient scenarios.
Our approach aims to deliver personalized health advice while minimizing medication risks. By leveraging multiple Large Language Models (LLMs), including ChatGPT-3, ChatGPT-3.5 Turbo, LLaMA-3.1-8B, LLaMA-3.2-11B-Vision, Meditron-7B, and Mistral-7B, we enable deep learning-based interpretation of multimodal inputs and generate human-like responses. The system processes vast amounts of medical data, extracts contextual insights, and delivers highly accurate, personalized healthcare recommendations.
- Multimodal Input Support: Accepts text, PDFs, scanned documents, and voice inputs for comprehensive healthcare guidance.
- Fine-Tuned Medical LLMs: Uses advanced language models trained on medical datasets to detect medication interactions and provide precise recommendations.
- Real-Time Adaptability: Continuously updates with the latest medical guidelines and user data to ensure accurate, context-aware healthcare assistance.
This intelligent Health LLM System bridges the gap between AI-driven healthcare solutions and real-world patient needs. It enhances accessibility, ensures patient safety, and delivers trusted medical insights, empowering individuals to manage their health more effectively.
- Multi-Modal Data Processing: Accepts text, PDFs, medical images, and voice inputs.
- AI-Driven Medication Guidance: Identifies potential drug interactions and provides safe usage recommendations.
- Personalized Assessments: Adapts based on patient history and health records.
- Fine-Tuned Medical LLMs: Optimized models for pharmaceutical and clinical tasks.
- High Accuracy (95.7%): Outperforms existing AI-driven medical models.
- Multi-Model Integration: Uses Evallms, and other top current LLM's for evaluation and scoring.
- ChatGPT-3 & ChatGPT-3 Turbo (for general medical Q&A)
- LLaMA 3.1 8B (for general text responses)
- LLaMA 3.2 11B Vision (for image-based diagnosis)
- Meditron-7B (highest accuracy model for medical NLP tasks)
- Mistral (for efficient real-time responses)
- OpenAI Whisper (for real-time voice to text conversion)
- EvalLM & GPT-4o (benchmark for performance comparison and accuracy)
- A to Z Medicines
- 250k Medicines Usage, Side Effects and Substitutes
- MIMIC-IV
- MIMIC-IV-ED
- MIMIC-IV-on-FHIR
- MIMIC-IV-Ext-BHC
- MIMIC-CXR
# Clone the repository
git clone https://github.com/Vignesh010101/Intelligent-Health-LLM.git
cd Intelligent-Health-LLM- Run the Model Inference file to infer the Intelligent Health LLM System directly from my fine-tuned model which are uploaded on HuggingFace platform.
- Run the MultiModal-Input-Output file to directly ask queries and you can also upload images & documents and get the responses.
- Fine-Tuned Model Accuracy: 95.7%
- Performance Gain over other existing top LLM models: +18.57%
- Response Time Optimization: Efficient real-time response with Meditron, LLaMA 3.2 Vision, Mistral and even GPT models.
- LLaMA 3.2 11B Vision Instruct Finetuned (GGUF)
- LLaMA 3.2 11B Vision Instruct Finetuned
- Meta-Llama-3.1-8B-finetuned
- Meditron-7b-finetuned
- Mistral-7B-Instruct-v0.3-finetuned
- More Models getting Fine-tuned on Medical Data
- Mobile App Integration (Coming-soon)
- Expanded Drug Interaction Database
- Enhanced Patient Data Privacy Mechanisms
This project is licensed under the Apache License