On the medical frontier: the 2006 KDD Cup competition and results.

The 2006 KDD Cup Competition featured three data mining tasks drawn from a medical imaging domain. At the core, all of these tasks were concerned with identifying pulmonary embolisms (PEs) from pre-processed computed tomography (CT) images of human lungs. However, these tasks were complicated by features such as multi-instance learning, stringent performance standards, hard-threshold evaluation functions, spatial correlations, and small training sets. This paper gives an overview of the data and medical imaging tasks, the competition and evaluation, and the competition victors.