Comparing different data fusion strategies for cancer classification

Abstract

Automatic cancer diagnosis can be performed based on different types of data sets. Some of them are microarray data, clinical trial data and cytopathological data. Usually class prediction is done by chosen classification method, for example a machine learning algorithm, and uses only one type of the available data. In this work an additional predictive value of a fusion of these three types of data is examined. To perform such research, authors upgrade and use their recently developed Spicy system. Different data fusion strategies have been tested on thyroid cancer data set. The workflow that has been created and the new module of a data fusion implemented in the Spicy system allows to qualify fusion of microarray data, clinical trials data and information about the Bethesda system class as a valuable method of prediction the thyroid nodule malignancy.