Syndrome Anomaly Detection Chapter 2 - NLP Protects Patient Privacy
In our previous post, we described that our aim is to be able to perform widespread disease monitoring to detect patterns of atypical disease presentation across communities. To do this the symptoms people are presenting with the need to be collected in an automatic, passive, and anonymous manner. This is where natural language processing (NLP) comes into play. Recently, specialized NLP models have demonstrated high levels of success to handle and interpret biomedical text, such as physicians' notes.
The system that we have designed has built upon that same application of biomedical NLP, however, trained to interpret patient-physician conversations. We have compacted our NLP model to be deployed entirely on a mobile device such as a cell phone. With this approach, we have the means to deploy a system that can automatically capture the symptoms each patient is presenting with while ensuring patient privacy, as the point of deployment is directly on the mobile device.
We had to ensure that the private conversations never leave the rooms. The NLP system is able to extract only the symptoms and provide that information exclusively to be analyzed to monitor for potential novel disease outbreaks. Our next goal is to determine how to make sense of the extract symptoms.