120 likes | 302 Views
Mobile Medicine . Pierfrancesco Belline Ivan Bruno Daniele Cenni Alice Fuzier Paolo Nesi Michela Paolucci. Semantic computing management for health care applications on desktop and mobile devices. Continuous Improvement in Healthcare.
E N D
Mobile Medicine Pierfrancesco Belline Ivan Bruno Daniele Cenni Alice Fuzier Paolo Nesi Michela Paolucci Semantic computing management for health care applications on desktop and mobile devices
Continuous Improvement in Healthcare • The ubiquity of mobile devices has presented new opportunities in distributing multimedia content • The medical community is continuously upgrading procedures and techniques. • Mobile Medicine integrates semantic computing with content distribution technologies in order to provide relevant content to HealthCare Providers. • during emergencies • for continuing education
System Architecture Monitoring & Reporting Content Production and Usage` AXMEDIS DRM Servers Grid Scheduler MobileMedicine Frontend IPR Management Node 1 User Management Node 2 Mobile Support Nodes ... Indexing/Querying Node N Recommendations Mobile Medicine CNP Databases Back Office Front-end Portal and Services Users
Suggestions (computed) • Computationally expensive because of – • Large N • Elements of N are complex • O(N^2) or worse if elements of N are complex • Estimation of similarity between elements using the following types of information • Users – static & dynamic information • Content – static (more so) & dynamic information • Descriptors will change over time based on users’ actions and will need to update the distances (similarities)
Additional Complexities... • Crossmedia descriptors can be very complex • Adding new elements requires the computation of many new similarity distances O(GM) Cross media content package. PDF HTML Flash
Clustering techniques • Clustering reduces the complexity of similarity distance calculation • K means (works only with numeric values) • assume K clusters • assume K centers (centroids) • assign each point to closest centroid • sum distances of each point • minimize that value by calculating new centroids* • reassign points to centroids • iteratively to convergence (clusters have minimal or no changes) • O(NKI) complexity • Select elements randomly from that cluster –or- • Do the calculation on a smaller set * vector calculus beyond scope of discussion and this presenter
Adding a new element... • Calculate a new element’s similarity to each cluster center. • Determine the cluster with minimum difference and place the element in that cluster. • Periodically recalculate clusters to optimize system with new elements.
What if your data is not numeric? • Distance can be the number of matched keywords • Distance based on keyword relationship in an ontology or taxonomy antimicrobial antiviral antibiotic An ontology is mapped onto a tree and distances are calculated based on the tree structure. oral oral intravenous intravenous
MobileMedicine Solution • MobileMedicine uses a new method of clustering which is hierarchical evolution of the k-medoids • k-medoids: similar to k-means, but each center point (medoid) is switched out until the lowest averaged distance is found, repeat until no medoid changes • hierarchical: an agglomerative or “bottom up” approach starts with each point as its own cluster, then begins to join points into clusters based on an increasing distance
Calculating Similarity DistancesD(U1:U2)=ksSxiSdi(U1:U2)+kdSyiDdi(U1:U2) T`s T`D • Sdi are static distance metrics: languages, age, continent/nationality, Groups, specializations • Ddi are dynamic distance metrics: set as friend, recommend, unfriend • ks and kd are weighting factors • Ts and Td are number of features to estimate similarity distance
Experiments • Different algorithms were used to cluster Content objects: • k-medoids, • hierarchical with complete linkage (farthest neighbor) • hierarchical with single linkage (nearest neighbor) • hierarchical with average linkage. • These clusters were then used to provide suggestions: C->U, C->G, and C->C
Conclusions • A solution has been presented to take advantage of semantic computing capabilities in order to give health care providers meaningful content on their mobile devices. • Questions???