- Location
- Virtual over Zoom
- Series/Type
- Bioethics Seminar, Biostatistics Seminar, DLSPH Event, Faculty/Staff Event
- Dates
- October 5, 2021 from 3:00pm to 4:00pm
Links
The Biostatistics Seminar Series presents:
“Clinical Natural Language Processing, An Application of Transformers and Rule-based Approaches to Deidentification” by Dr. Alistair Johnson, Hospital for Sick Children
Register here.
Abstract
Deidentification is the process of removing individual information (names, locations, dates, identifiers), and is an important task in the medical context as it protects patient confidentiality. In the context of natural language, deidentification is cast as a named entity recognition task, involving the location of spans of text which contain identifiable information within clinical text. We will review past approaches for deidentification through to recent models, including: rule-based approaches, manually engineered features combined with machine learning, recurrent neural networks, and transformer based models. Particular challenges in the annotation and evaluation of these models will be raised through case studies of real-world applications.