Open Source on github

MedLibreGPT
Enhancing Medical Data Management and Patient Care through AI and Open-Source Integration

Our goal is to elevate both medical data management and patient care by seamlessly integrating PrivateGPT with on-premises open-source storage.

Medical Knowledge Domain

We grant access to an extensive collection of medical literature on an on-premises server, meticulously curated to cater to specific medical domains.

AI Digital Twin Patient

We gather comprehensive patient data from a clinic’s on-premises Nextcloud server. By examining longitudinal data, ranging from surgical planning to long-term outcomes, we craft distinct risk profiles for digital twin patients.

AI Digital Twin Physician

We develop digital twins of physicians that mirror their knowledge and experience. This innovation allows for predictions of treatment risks, fostering safer and more customized medical decisions.

Case study

What does that mean in practical terms:

(1) We have a patient who lacks sufficient bone mass for an implant, necessitating bone augmentation. The patient requests a biological method for augmentation.

(2) The medical domain server contains an extensive library of literature on bone augmentation, including topics like Platelet-Rich Fibrin used as an Autogenous Graft.

(3) Our clinic’s server holds records of five successful longitudinal cases where patients received Platelet-Rich Fibrin. Utilizing this data, we generate a digital twin of the patient. Like an AI summary of patients.

(4) The skill level of the Digital Twin Physician is matched with that of the digital twin patient. Like giving a prompt an instruction about the skill level.

Although the physician has no prior experience with Platelet-Rich Fibrin as an Autogenous Graft, he is experienced in standard augmentation procedures. After conducting research in the specialized medical domain and analyzing longitudinal patient data, the physician decides to proceed with this new procedure.

This type of decision-making occurs every day, often on an unconscious level. Our aim is to employ AI to aid in this process.

Prior Research
We were the first to develop and implement augmented reality-based teleconsultation in image-guided surgery. We introduced the term “Knowledge Guided Surgery.”


FWF Research Project P 12464
The optical interface for augmented reality in computer-assisted navigation and 3D visualization in surgery

Google Scholar

Github

https://github.com/AI-in-Health/MedLLMsPracticalGuide

Links

https://watson.smile.wien

http://medlibre.com

Sample Questionaires

Beispiel 1

Beispiel 2

Chat

01

Open Source

MedLibre for free software in medicine

02

Private Cloud

Nextcloud as open source private cloud

03

PrivateGPT

PrivateGPT for sensitive patient data

Medlibre Dual-Mode AI

MedlibreGPT integrates a dual-mode concept, combining a secure, local PrivateGPT mode for processing sensitive patient data with a connected mode for interfacing with OpenAI’s API. The PrivateGPT mode operates within a local environment, ensuring high-level data confidentiality, tailored for use-cases requiring stringent privacy controls. In contrast, the OpenAI API mode expands the system’s capabilities, offering broader functionality while adhering to robust security protocols. This dual-mode architecture allows us to seamlessly switch between local data protection and advanced AI features, providing a versatile and secure solution for managing sensitive data in AI-driven applications.

Specialized Domains

MedlibreGPT uses fine-tuned local Large Language Models (LLMs) for specific medical domains, surpassing the capabilities of generic models like ChatGPT. Our methodology involves a comprehensive processing of available medical literature, combined with real-world clinical insights from healthcare professionals. Key to our approach is the integration of extensive medical knowledge with patient histories stored securely in our Nextcloud private cloud. This synthesis aims to significantly improve the precision and relevance of AI applications in medical diagnostics and treatment planning. By enriching these local LLMs with both broad medical information and individual patient data, we aim to advance healthcare decision-making and outcomes.

Local for Patient Data

MedlibreGPT is specifically engineered to securely handle sensitive patient data, ensuring its confinement within our system. Patient information is safely stored in our Nextcloud private cloud. Specialised Large Language Models (LLM), tailored for very narrow medical domains, operate exclusively on our server, enhancing data privacy and security. This open source system is HIPAA compliant and scalable to any size using the Nextcloud federated server concept. It offers deployment flexibility, accommodating in-house Virtual Machine (VM) setups or external hosting with a secure VPN connection. This provides a robust and adaptable solution for maintaining the utmost confidentiality and integrity of patient data in healthcare applications.