Prototypical Networks for Interpretable Diagnosis Prediction
The use of deep neural models for diagnosis prediction from clinical text has shown promising results. However, in clinical practice, such models must not only be accurate but provide doctors with interpretable and helpful results. In this talk, we introduce ProtoPatient, a novel method based on prototypical networks and label-wise attention with both of these abilities. ProtoPatient makes diagnosis predictions based on parts of the text that are similar to prototypical patients–providing justifications that doctors understand. Our results demonstrate that the model makes state-of-the-art predictions and provides valuable explanations for clinical decision support.
Betty van Aken
Applied Research Scientist / Guest Researcher Grammarly / Berlin University of Applied Sciences and Technology (BHT)
Betty van Aken has conducted her Ph.D. research at the DATEXIS research group of the Berlin University of Applied Sciences and Technology. Since 2018, her research has focused on deep learning approaches for natural language processing. In this regard, her previous work involved adapting and explaining deep neural networks for specialized domains. Her most recent focus is the automatic analysis of clinical notes with the goal of supporting doctors in the process of differential diagnosis.
Since January 2023, Betty works as an Applied Research Scientist at Grammarly.