Application of Nonnegative Tucker Decomposition in Medical Data Analysis

Abstract

In recent years, we have been studied about medical data analysis, especially for the rehabilitation data provided by a hospital, and extracted the recovery tendency of patients from their Functional Independence Measure (FIM) data. This time, we adopt the nonnegative Tucker decomposition (NTD) method, which is known as an extension of the nonnegative matrix factorization (NMF) to higher-dimensional data, to the medical data built up by piling each FIM data at some time points for several patients. Since the all elements of the tensor and matrices obtained by the NTD are nonnegative, it is expected that this method makes the interpretation of the characteristic vectors which are obtained from the resulting matrices easy and intelligible in comparison with our former approach, which used the multi-dimensional principal component analysis (MPCA). The experimental results show the effectiveness of proposed approach.