On knowledge-based improvement of biomedical pattern recognition-a case study

Abstract

Most biomedical pattern recognition (BPR) systems use the classical statistical pattern recognition strategy in which a feature hyperspace is constructed for the problem followed by a statistical discriminant analysis procedure. It is shown that there are several essential drawbacks with this conventional approach. These limitations often lead to highly erroneous classification results subject to time-consuming interactive corrections in existing systems. To overcome such limitations, a pilot study on applying knowledge-based techniques to chromosome classification has been carried out. The scheme developed as a result of this case study is described in detail. The system has been implemented and tested on metaphase imagery in preliminary experiments, the results of which are also presented.<>