A Mixture of Views Network with Applications to CADx Systems

This work examines data fusion methods for multi-view data classification. We present a decision concept

which explicitly takes into account the input multi-view structure, where for each case there is a different subset of relevant

views. This data fusion concept, which we dub Mixture of Views, is implemented by a special purpose neural network

architecture. The single view decisions are combined by a data driven decision, according to the relevance of each view in a

given case, into a global decision. The method was applied to two challenging computer-aided diagnosis (CADx) tasks. First, it is

demonstrated on the task of classifying breast microcalcifications as benign or malignant based on CC and MLO mammography

views. Additionally, the method was utilized to segment Multiple Sclerosis (MS) white matter lesions. The experimental results

show that our method outperforms previously suggested fusion methods.

* M.Sc. research supervised by Prof. Jacob Goldberger