iheallab/apricotM
📦 Open Source Projectiheallab
A deep learning framework using state-space models for real-time patient acuity prediction in intensive care units.
ApricotM represents a significant advancement in clinical decision support systems by integrating state-space models (SSMs) to analyze high-frequency EHR data. Unlike traditional recurrent neural networks or standard transformers that can be computationally expensive for long-sequence clinical data, ApricotM utilizes the Mamba architecture to maintain high performance with linear scaling. The framework is designed to ingest multi-modal EHR inputs, including vital signs and laboratory results, to output real-time acuity scores. This approach allows for the continuous assessment of patient stability, helping medical staff anticipate deterioration and optimize therapy allocation. The repository includes the complete pipeline for data preprocessing, model training, and inference, specifically optimized for the unique temporal dynamics found in ICU settings.
💡Highlights
- ├─State-space model for EHR data
- ├─Real-time ICU acuity prediction
- └─Linear scaling with Mamba architecture
🎯For
- ├─Healthcare Data Scientists
- └─Clinical Researchers