Induction of image retrieval knowledge from radiologists' reading instances

This paper proposes a Multi-Decision-Tree Induction (MDTI) approach to image prehanging and discusses how it can facilitate knowledge acquisition and maintenance through the induction of knowledge embedded in radiological image reading cases which have the characteristics of inconsistent retrievals, incomplete input information, and multiple decision outcome classes. We present empirical comparisons of the MDTI approach with Backpropagation network algorithm, and the traditional knowledge acquisition approach using the same set of cases in terms of the recall rate, the precision rate, the average number of prior examinations suggested, understandability of the acquired knowledge, and the required learning time. The results show that the MDTI approach outperforms the Backpropagation network algorithm in all performance measures studied.