A visualization oriented data mining tool for biomedical images

摘要

We present a visualization oriented data mining tool designed for biomedical images with focus objects. The main functions of the tool are accessible via two interfaces: Query by Example and Query by Features. Each interface has visual feedback functions associated with it. The query-by-example interface allows the user to do similarity search using multiple image features. The visual feedback consists of positions of feature values in the database population distribution. Classification incorporates a measure of uncertainty that is based on class differentiation quality. Visual feedback includes visualizations of class likelihoods and uncertainty. Query-by-features interface helps the user explore the relationships among features and diagnoses. The visual feedback includes example images of extreme feature values, images from the database that satisfy query criteria and diagnoses distribution charts. Overall, the tool is designed to increase awareness of visual indicators of health risk in biomedical images. While the present application is focused on skin cancer awareness, the ideas are applicable to other types of biomedical images with focus objects, such as breast tumor and wound images.