Medical Concept Extraction and Relationship Classification from Patient Records

Tech Stack: Python, Pandas, PyTorch, FlairNLP, Scikit-learn

GitHub: Link

• Created ontology-aware NER and Relationship classification models by implementing a novel UMLS-based data augmentation technique.
• Developed BiLSTM-CRF & Transformer-CRF models for entity extraction and a BiLSTM model for relation classification from patient records.
• Achieved a state-of-the-art 91% Micro F1-score for entity recognition with 85% accuracy for the relation classification on Harvard Med's "n2c2 adverse drug events (ADE) and medication extraction in the electronic health records" dataset.