DWT-PCA Image Fusion Technique to Improve Segmentation Accuracy in Brain Tumor Analysis

Abstract

Because of its high clinical significance and varied modalities; magnetic resonance (MR) imaging procedures are widely adopted in medical discipline to record the abnormalities arising in a variety of internal organs of human body. Each modality of the MRI, such as T1, T2, T2C, Flair, and DW has its own merit and demerits. Hence, in the proposed work, a unique computer-assisted technique (CAT) is proposed to evaluate the abnormalities in MR images, irrespective of its modalities. Proposed CAT has the following stages: (i) Discrete Wavelet Transform Based Principal Component Averaging (DWT-PCA) image fusion, (ii) Tri-level thresholding based on Social Group Optimization and Shannon’s entropy, and (iii) Watershed segmentation. This approach is experimentally assessed with MICCAI brain cancer segmentation (BRATS 2013) challenge database. Experimental results confirms that the proposed approach is efficient in offering better values of Jaccard (84.33%), Dice (90.86%), sensitivity (99.93%), specificity (90.67%), and accuracy (95.74%) compared with the single modality registered brain MR images. Hence, the proposed work is extremely significant for the segmentation of abnormal region from the brain MR images registered using Flair, T1C, and T2 modalities.