Artificial Prediction Markets for Lymph Node Detection

Prediction markets are forums aimed at predicting the outcome of future events of interest such as election results. People participate in a prediction market by buying contracts on the possible outcomes. They are rewarded after the outcome is known based on the number of contracts purchased for the correct outcome. The Artificial Prediction Market is a novel machine learning method that simulates a prediction market where the participants are trained classifiers instead of people. In this work we present an application of the Artificial Prediction Market to lymph node detection from CT images. An evaluation on 54 CT volumes shows that the detector trained with the Artificial Prediction Market has a detection rate of 81.2% at 3 false positives per volume, while an Adaboost classifier trained on the same features obtains a detection rate of 79.6% at the same false positive rate.