Application of Convolutional Neural Networks and Rolling Guidance Filter in Image Fusion for Detecting Brain Tumors

Medical image fusion is the technique of integrating images from several medical imaging modalities without causing any distortion or information loss. By preserving every feature in the fused image, it increases the value of medical imaging for the diagnosis and treatment of medical conditions. A novel fusion mechanism for multimodal image data sets is proposed in this paper. Each of the source image is smoothened using cross guided filter in the initial step. Guided filter output is further smoothened to remove fine structures using rolling guidance filter. Then details (high frequency) of each source image are extracted by subtracting the rolling guidance filter output from corresponding source image. These details are fed to convolutional neural networks to obtain decision maps. Finally the source images are fused based on decision map using maximum rule of combination. We assessed the performance of our suggested methodology using several pairs of medical imaging datasets that are accessible to the general public. According to the quantitative evaluation, the recommended fusion strategy for multimodal image fusion improves the average IE by 12.4%, MI by 41.8%, SF by 21.4%, SD by 22.81%, MSSIM by 31.1%, and by 39% when compared to existing methods, which makes it appropriate for use in the medical field for accurate diagnosis.