ERROR ENTROPY MINIMIZATION FOR BRAIN IMAGE REGISTRATION USING HILBERT-HUANG TRANSFORM AND ECHO STATE NEURAL NETWORK

Rajeswari R, Anthony Irudhayaraj A, Purushothaman S

Abstract: This paper presents the work on functional magnetic resonance imaging slice registration. Hilbert-Huang transform is used to extract the features of the source image slice and target image slice. The features are used as inputs for the echo state neural network (ESNN) which is recurrent neural network. The training of ESNN is carried out by changing the different number of reservoirs in the hidden layer that result in minimum error between floating and target image slice after registration. Statistical features of the decomposed signals of the source and target image slices are used in the input layer of ESNN. The accuracy of the registration is high.

Keywords: Echo state neural network (ESNN), Empirical Mode Decomposition (EMD), Hilbert Transform (HT), Hilbert-Huang transform (HHT), Image alignment, Medical image registration, Processing element

DOI: https://doi.org/10.15623/ijret.2014.0319156