AI for Healthcare
Dev & Doc
A Blueprint for De-identification of Patient Records | Dev&Doc #09

A Blueprint for De-identification of Patient Records | Dev&Doc #09

How to deploy large language models to de-identify patient records and enable research and commercial use.

No transcript...

You can also listen to the episode on YouTube DevAndDoc and Spotify DevAndDoc

🎙️ Podcast Highlights

  • Introduction: An in-depth exploration of the challenges and solutions in de-identifying hospital records.

  • De-identification Approaches: Discussing various strategies used in the anonymization of hospital data.

  • NER Episode Card: Insight into Named Entity Recognition (NER) and its applications in this context.

  • Over-Redaction Issues: The problems associated with excessive anonymization and data loss.

  • Building High Recall Models: Sharing experiences in creating models with over 99% recall.

  • The Art of Annotations: Strategies for developing high-performing models through effective annotation.

  • Dev and Docs Annotation Method: Exploring the annotation methodology proposed by Zeljko et al.

  • Performance in New Environments: Ensuring model efficacy in different hospitals or healthcare settings.

  • Looking to the Future: Anticipating future developments and challenges in this field.

  • Synthetic Data: The role and potential of synthetic data in improving and testing de-identification methods.

Thank you for listening,


AI for Healthcare
Dev & Doc
Our podcast is structured as a collection of conversations between doctors and developers exploring the potential of AI in healthcare. Only the last couple of episodes are on Substack, you can find the older ones on YouTube or Spotify -