Predicting and Optimizing Classifier Utility with the Power Law

Abstract

Data Modeling is an essential first step for data preparation in any data mining procedure. Conventional entity-relational (E-R) data modeling is lossy, irreproducible, and time- consuming especially when dealing with unstructured image data associated with complex systems like the human brain. We propose a methodological framework for more objective E-R data modeling by eliminating the structured content- dependent metadata associated with the unstructured data. The proposed method is applied to epilepsy-related image data and a system called the human brain image database system (HBIDS) is developed accordingly. Supported with navigation, segmentation, data fusion, and feature extraction modules, HBIDS provides a content-based support environment (C-BASE). Such an environment potentially provides an unlimited (ad hoc) query support with a reproducible and efficient database schema. Switching between different modalities of data, while confining the feature extractors within the object(s) of interest, HBIDS yields anatomically specific query results. The price of such scheme is large storage requirements and relatively high computational cost. Examples of navigation through unstructured image data and content-based retrieval are presented in this paper. The results show the potential of HBIDS in content-based data management for decision support systems in real life medical applications.