Interactive Clinical Query Derivation and Evaluation

For an effective search and management of large amounts of medical image and patient data, it is relevant to know the kind of information the clinicians search for. This information is typically represented in the search queries of the clinicians, which they send to retrieve the related text and images. Collecting these queries via typical expert interviews, however, is inappropriate. The reason is that for a successful communication during the interview some medical knowledge background of the knowledge engineer becomes necessary, which is usually not available. Therefore, alternative techniques are required to obtain relevant information about clinical search queries that are independent of the expert interviews. The query pattern derivation approach described here is one technique to gain this information. It is based on the prediction of clinical query patterns given domain ontologies and corpora. The patterns identified in this way are then presented to the clinical experts via an interactive browser for knowledge elicitation and evaluation purposes. Being an interactive tool, the Clinical Query Pattern Browser also supports the communication process between the clinical expert and the knowledge engineer. transfer from the clinical expert to the knowledge engineer. This was due to the reason that medical knowledge is too specific and too sensitive for a knowledge engineer to come up with the appropriate interview questions and to be able to process the interview answers in an adequate manner. The clinical experts and the knowledge engineers have a quite different perception of the medical domain; the former mostly focuses on special cases and outstanding details, whereas the latter's focus is to generalize or to abstract the domain of interest. To overcome the difficulties in the communication process between the clinical experts and knowledge engineers, our aim is to establish tools and methodologies that support the knowledge elicitation process. In particular, we are interested in identifying queries, which are used by clinicians to retrieve medical images and related patient data and which we were not able to identify during the expert interviews. To achieve this goal, we follow two steps: Firstly, we attempt to predict the search queries of the clinical experts. This is done based on our query pattern mining approach that uses domain