Mammographic information analysis through association-rule mining
Xiaozheng Wang, Michael R. Smith, Rangaraj Mandayam Rangayyan · 2004
The increasing availability of large clinical and biomedical data repositories provides researchers with substantial opportunities for data analysis and knowledge discovery. Data mining is an expanding research frontier that provides numerous efficient and scalable methods to extract patterns of interest in datasets. The University of Calgary Atlas of Mammograms (U of C Atlas) contains digital mammographic images and textual reports of radiologists acquired from Screen Test Alberta. Many advanced image-processing techniques have been applied to the images in this dataset. However, research has not been conducted to take advantage of data-mining techniques, which motivates us to investigate the functionality of association-rule mining techniques to discover patterns of interest in the existing dataset. This paper describes preliminary results of the application of applying association-rule mining techniques to the U of C Atlas. We propose a new breast mass classification method based on quantitative association-rule mining. The experiments conducted on the U of C Atlas show that many interesting rules can be generated from this dataset, and indicate previously unobserved patterns in the information contained in the atlas.