Patient classification using association mining of clinical images

Automated clinical image data collection tools and apparatus are becoming increasingly important to the medical industry, and imaging databases are growing at an unprecedented rate. Consequently, grid-based telemedicine efforts require the autonomous classification of patient images from distributed sources for fast and accurate image storage, management, and retrieval. In this paper, we present a unique algorithm that performs feature discovery to find class-wise isomorphic association rules (ARs) among features. By discovering ARs, we are able to find unique and useful knowledge in images. To find knowledge, we first uniformly segment every image in a series and extract color and texture features for every segment., Next, we discover ARs for the color and texture features for image segments. We then exploit redundancy in the differentials of rule sets for the autonomous classification of patient image data with significant sensitivity and specificity. We demonstrate the efficacy of our approach with experimental results on a data set of diabetic retinopathy patients.