Remember that some weeks ago we were talking about how we approach de-identification at Konplik? We published a series of posts describing the relevance of de-identification technology in allowing medical data to be shared free of Personal Health Information (PHI). In these posts, we also described the methodology we employ at Konplik to develop and maintain our AI models for this task and some notes on the way we implement them. Now, you can test it yourself using this demo we have published online at Konplik Demos.
This online demo allows you to test the basic model available at Konplik for de-identification purposes in English. This technology can be used to anonymize your EHRs or, typically, any medical report. It is a deep learning-based model that we have built using the BioClinicalBERT pre-trained model and then fine-tuned for the sequence labeling task. The training carried out was based on collections of medical records.
Please, take a look at our demo, conduct your tests and, of course, we will be more than happy to receive any feedback you may have.