Multi-modal Medical Image Fusion Technique to Improve Glioma Classification Accuracy

Usually, the low grade (LGG) and high-grade glioma (HGG) classification algorithms proposed in the literature directly concatenate the magnetic resonance image (MRI) modalities that directly affect the accuracy and precision results. Therefore, here this problem is highlighted at the initial level by applying a multi-modality fusion scheme. First, multi-level edge-preserving filtering (MLEPF) is applied to decompose the source images into fine-structure (FS), coarse-structure (CS), and base (BS) layers. Then, the novel sum-modified Laplacian (NSML) and parameter Adaptive PCNN based fusion strategy is adopted for the fusion FS and CS layers. Where region energy (RE) and information entropy (IE) based fuzzy pixel rules are implemented for the fusion of BS layers. The final fused image is achieved by integrating all the three fused layers. Visual and quantitative analysis prove that the proposed scheme results are satisfactory compare to the state-of-art. The results were also evaluated by using the Google Inception V3 convolutional neural network (CNN).