A Mobile Recommender System for Location-Aware Telemedical Diagnostics

As recommender systems have proven their effectiveness in providing personalised recommendations based on previous user preferences in e-commerce, this approach is to be transferred for use in medicine. In particular, the aim is to complement the diagnoses made by physicians in rural hospitals of developing countries, in remote areas or in situations of uncertainty by machine recommendations that draw on large bases of expert knowledge to reduce the risk to patients. To this end, a database of patients’ medical history and a cluster model is maintained centrally. The model is constructed incrementally by a combination of collaborative and knowledge-based filtering, employing a weighted similarity distance specifically derived for medical knowledge. In the course of this process, the model permanently widens its base of knowledge on a medical area given. To give a recommendation, the model’s cluster best matching the diagnostic pattern of a considered patient is sought. Fuzzy sets are employed to cope with possible confusion in decision making, which may occur when large data sets cause clusters to overlap. The degrees of membership to these fuzzy sets are expressed by the Mahalanobis distance, whose weights are derived from risk factors identified by experts. The therapy actually applied after the recommendation and its subsequently observed consequences are fed back for model updating. Readily available mobile digital accessories can be used for remote data entry and recommendation display as well as for communication with the central site. The approach is validated in the area of obstetrics and gynecology.