ABSTRACT
This chapter discusses the image mining methodology, from spatial analysis of millions of cells in virtually multiplexed studies using cell density heatmaps, to correlating potentially relevant patterns with clinical observations by optimizing predictive values and survival time prediction. Image mining describes the process of generating knowledge from information implicitly available in images. When focusing on applications in the field of histopathology, image mining of collections of digital whole tissue slides, generated from human patient samples with known disease progression, enables the extraction of diagnostic knowledge. Image mining requires the aggregation of all measurements for a single patient into a single feature vector. This vector comprises various numerical and categorical entries, which provide the input for the subsequent data mining methods. The chapter also discuss regions of interests, which are generated by expert pathologists, and next regions, which are detected automatically using high-level image analysis.
