A hybrid fuzzy-wavelet approach to medical image fusion

Much concern remains in the ability of physicians to provide a timely and accurate medical diagnosis. Tremendous efforts are made in the thorough analysis of images to extract, combine and visualize the most representative features within an image for the characterization of the respective pathological condition. The aim of this research work is to provide an alternative approach for effectively combining relevant anatomical and functional information from different scanning modalities in a fused output image.

For this work, preprocessed and coregistered Magnetic Resonance (MRI) images and Positron Emission Tomography (PET) scans are decomposed by a discrete wavelet transform (DWT) in successive resolution steps and represented by matrices of wavelet coefficients at the corresponding resolution levels. These MRI and PET coefficient matrices are then fused in a fuzzy inference system (FIS) to produce a .single set of coefficient matrices which characterize both the PET and MRI features. The "fuzzy engine" implementation allows for a robust and flexible representation of the rules that determine the method of coefficient selection for the actual fusion of the wavelet coefficients. Finally, the resulting fused coefficient matrices are reconstructed by an inverse wavelet transform (IWT) in order to obtain the output fused image.

The fuzzy-wavelet system hereby proposed consists of a novel fusion technique by taking advantage of the superb spatial and frequency characterization of wavelets and of the robustness of the fuzzy inference system design. Coregistered sets of brain images are analyzed by using different wavelet basis functions and different FIS designs, and all resulting images demonstrate the effectiveness of the method. Best results are obtained by utilizing Daubechies and biorthogonal wavelets with lower vanishing moments; FIS designs may be optimized to enhance specific features from either PET or MRI input images at the physician's criteria. The output fused image may be further compressed by processing with the DWT for ease of storage.