An Agent Supporting Symptom Elicitation in Physician-Patient Dialogue

Abstract

The main objective of the framework we are proposing is to help the physician obtain information about the patient’s condition in order to reach the correct diagnosis as soon as possible. In our proposal, the number of interactions between the physician and the patient is reduced to a strict minimum on the one hand and, on the other hand, it is made possible to increase the number of questions to be asked if the uncertainty about the diagnosis persists. These advantages are due to the fact that (i) we implement a reasoning component that allows us to predict a symptom from another symptom without explicitly asking the patient, (ii) we consider non-binary values for the weights associated to the symptoms, and (iii) we introduce a dataset filtering process in order to choose which partition should be used with respect to some particular characteristics of the patient The experimental results we obtained are very encouraging.