Similarity retrieval of cardiac reports

Abstract

Mining medical reports can reveal important information correlating diagnosis with raw measurements helping in decision support. In this paper we address the problem of finding similar measurement reports for aiding clinical decision support. Specifically, we present a new approach to generating document class models for measurement reports using a multi-scale feature-value kernel. The class models serve as natural feature selection mechanism as well as indexes to large report collections. A document retrieval algorithm based on document class models is presented to enable similarity retrieval of pre-diagnosed reports. Collaborative filtering-guided assembly of associated disease labels is used to achieve clinical decision support.