Quantitative Assessment of Estimation Approaches for Mining over Incomplete Data in Complex Biomedical Spaces: A Case Study on Cerebral Aneurysms

摘要

Biomedical data sources are typically compromised by fragmented data records. This incompleteness of data reduces the confidence gained from the application of mining algorithms. In this paper an approach to approximate missing data items is presented, which enables data mining processes to be applied on a larger data set. The proposed framework is based on a case-based reasoning infrastructure which is used to identify those data entries that are more appropriate to support the approximation of missing values. Moreover, the framework is evaluated in the context of a complex biomedical domain: intracranial cerebral anenrvsms. The dataset used includes a wide diversity of advanced features obtained from clinical data, morphological analysis, and hemodynamic simulations. The best feature estimations achieved errors of only 7%. There are, however, large differences between the estimation accuracy achieved with different features.