Sepsis Risk Predictor Project: An Open Source Platform for Clinical Data Integration and AI-driven Sepsis Detection
- Role: Project Lead/Senior Data Scientist
- Start Date: Feb 1, 2022
- End Date: Dec 31, 2025
- Status: Ongoing
- Institution: Cross Platform Open Source Project.
- Funding Institution: Open Source-funded project.
- Website:
Summary
FlexiCare: Framework for Processing Clinical Data for Sepsis Risk Prediction
Processing clinical notes is challenging due to their lexical and semantic diversity. FlexiCare addresses this by using natural language processing to create numerical structures for clinical notes, maintaining context and meaning. The Word2Vec algorithm, combined with a convolutional neural network (CNN), detects sepsis based on the SOFA formula. Using Python, data from the MIMIC-III database is structured and filtered for analysis. LightGBM processes structured data with 86% accuracy, while the full model (Word2Vec + CNN) achieves 89.45% accuracy in modeling patient behavior from clinical notes.
Co-researchers
- Prof. MSc. Vivian Milen Orejuela R.
- Prof. PhD. John Alexander Sanabria O.
Academic Events
- CARLA 2022. A Machine Learning-Based Missing Data Imputation with FHIR Interoperability Approach in Sepsis Prediction. Porto Alegre, Brasil.
Project Images
