A new LabAI menu option is in the process being added to OpenELIS to integrate AI-powered functionalities into the system. This is implemented using a Liquibase database change set, ensuring that the menu item is only added if it does not already exist.
Questions for Mentors Regarding AI Model Integration in OpenELIS
As you work on integrating AI into OpenELIS, would like your guidance on selecting the most suitable models for different functionalities. Based on our use cases, here are some considerations:
1. Predictive Analytics & Diagnostics
- Would machine learning models like Random Forest, XGBoost, or LightGBM be ideal for predicting lab test results and detecting anomalies?
- If we plan to analyze medical images (e.g., blood smears, pathology slides), would a CNN-based deep learning model be a viable approach?
2. Automated Result Validation
- For identifying irregularities in test results, should we consider anomaly detection models like Autoencoders or Isolation Forests?
3. Natural Language Processing (NLP) for Clinical Data
- Should we leverage BioBERT, ClinicalBERT, or Med7 for extracting relevant information from lab reports, prescriptions, and doctor’s notes?
- Would a GPT-based model be useful for automating report generation and summarization?
4. Disease Surveillance & Outbreak Prediction
- For tracking and predicting outbreaks, should we integrate time-series models like LSTMs or epidemiological models (SEIR)?
- Could Graph Neural Networks (GNNs) help in mapping disease spread across regions?
5. Workflow Optimization & Automation
- Would reinforcement learning (RL) be effective in optimizing lab processes, such as sample processing and resource allocation?
Also
- Do you recommend a specific approach or model for each use case?
- Are there existing AI models in the healthcare domain that OpenELIS could adopt or fine-tune?
- What challenges should contributors anticipate in integrating AI with current database and infrastructure?
Your insights will help in refine the AI strategy and implementation roadmap. Looking forward to your guidance! 🚀