Medical NER with BERT
A fine-tuned BERT model that extracts medical entities from clinical text and turns them into structured JSON, reaching 96% accuracy, precision, recall and F1.
Highlights
- Fine-tuned a BERT-based named entity recognition model on clinical text
- Designed an entity-level inference pipeline that converts token predictions into structured medical JSON
- Implemented BIO tagging, token-label alignment, padding, attention masks and a train/test split
- Reached 96% accuracy, precision, recall and F1-score
Overview
Clinical notes are written for people, not systems. This project fine-tunes BERT to pick out medical entities from that text and outputs them as structured data that other tools can use.