CORe-clinical-diagnosis-prediction
| Property | Value |
|---|---|
| Author | DATEXIS |
| Paper | Clinical Outcome Predictions from Admission Notes using Self-Supervised Knowledge Integration |
| Framework | PyTorch |
| Task | Clinical Diagnosis Prediction |
What is CORe-clinical-diagnosis-prediction?
CORe (Clinical Outcome Representations) is an advanced medical AI model built upon BioBERT, specifically designed for predicting clinical diagnoses from patient admission notes. The model has undergone specialized Clinical Outcome Pre-Training using clinical notes, disease descriptions, and medical articles, making it particularly adept at understanding medical context and terminology.
Implementation Details
The model implements a multi-label classification approach, supporting predictions across 9,237 different labels, including both 3- and 4-digit ICD9 codes and their textual descriptions. It utilizes the transformers library and can be easily integrated into existing workflows using PyTorch.
- Built on BioBERT architecture with specialized medical training
- Supports multi-label ICD9 code prediction
- Implements hierarchical code structure (3-digit and 4-digit ICD9 codes)
- Uses transformer-based architecture for optimal text understanding
Core Capabilities
- Processing of clinical admission notes
- Multi-label diagnosis prediction
- ICD9 code classification
- Integration of hierarchical medical knowledge
- Support for both code and textual description outputs
Frequently Asked Questions
Q: What makes this model unique?
The model's unique strength lies in its specialized Clinical Outcome Pre-Training and ability to handle hierarchical ICD9 codes while incorporating both numerical codes and textual descriptions. This makes it particularly effective for real-world clinical applications.
Q: What are the recommended use cases?
The model is best suited for automated diagnosis coding from clinical admission notes, particularly in hospital settings where ICD9 code assignment is required. It's recommended to use the 3-digit code predictions for optimal performance, as these have been thoroughly evaluated in the original research.
