Self-Organizing-Map-Based Detection on Hematopoietic Tumors in a Nonstationary Environment

This paper proposes a scheme of detecting the screening data of hematopoietic tumor patients, using block-matching-based self-organizing maps. The data of an examinee frequently lacks several of the item values, and hence the data is presented to a map after averages of non-missing item values are substituted for items with no values. It is basically judged, by observing the label of a winner block in a map, whether the data presented to the map belongs to the class of hematopoietic tumors. Proposed scheme allows us to construct maps not only in stationary environments where members in a training data set never change but also in nonstationary environments where the data set is suddenly updated during learning. Simulation experiments established that the proposed scheme achieves high accuracy of correctly classifying the data, even if the map is in nonstationary environments.