Study of similarity measures for case-based reasoning in transcatheter aortic valve implantation

Case-Based Reasoning (CBR) uses previous experiences to solve similar current problems. The basic hypothesis is that similar cases should have similar solutions. In the case of Transcatheter Aortic Valve Implantation (TAVI), the CBR could help practitioners to plan the procedure. Four steps compose a CBR: retrieve, reuse, revise and retain. Defining a convenient similarity measure (SM) is essential in the retrieve step. This study aims to analyze the performance of different similarity measures and attribute selections. Generally in the retrieve step, a standard weighted heterogeneous similarity measure (WHSM) is used, in association with the k-nearest neighbor algorithm. Based on WHSM, we considered new definitions of SMs dedicated to decision support for TAVI. They include attributes selection and weight determination through a clinical decision tree. The performance of SMs was evaluated on a set of 100 cases with a leave-one-out cross validation. Results show that the CBR retrieving process can be improved by using dedicated SMs.