Multimodality medical image fusion using non-subsampled rotated wavelet transform for cancer treatment

Abstract

This paper presents non-subsampled rotated wavelet transform based feature extraction approach to multimodality medical image fusion. The non-subsampled rotated wavelet filters are designed to extract textural and edge features from axial brain images from the modalities viz. computed tomography and magnetic resonance imaging. These extracted features are selected using entropy-based fusion rule to form composite spectral plane. Entropy-based fusion rule preserves dominant spectral features and imparts all relevant information from both the modalities to the fused image. The fused image is reconstructed using inverse transform from the composite spectral slice. The proposed algorithm is evaluated using 39 sets of pilot images of 23 patients subjectively and objectively. Three expert radiologists have verified the subjective quality of the fused images to ascertain abnormality. Subjective and objective evaluation reveal that the fused images using proposed algorithm are superior in terms of visualisation of abnormalities over the state of the art wavelet based algorithms.