A Novel Multi-Level Medical Image Fusion Using Ordered Pixel Mean Filter and Parameter Adaptive Dual Channel Weighted Pulse Coupled Neural Networks

Medical image fusion is a crucial technique that has been extensively employed in clinical diagnosis and treatment planning, driven by rapid advances in imaging techniques. Though several methods have been introduced in the past for effective fusion of source modalities, there is still room to improve fusion accuracy. In this work, a novel method of image decomposition is introduced using the mean of ordered pixel values filter, and is applied for multi-scale decomposition of input images for fusion. Further, a weight computation mechanism based on singular values of the Singular Value Decomposition (SVD) method is introduced to fuse coarse layers of source images, while the detail layers are fused at multiple levels with the help of Parameter-Adaptive Dual-Channel Weighted Pulse Coupled Neural Networks (PADCWPCNN). The fusion performance of the proposed method is evaluated on 125 slices of diverse brain diseases using both subjective and objective criteria. Experimental results demonstrated the efficacy of the proposed method, as it improved Average Gradient (AG), Standard Deviation (SD), Entropy (H), Spatial Frequency (SF), Edge Strength (QABF), and Mutual Information (MI) by 57%, 3%, 14.7%, 17.9%, 46%, and 14%, respectively, making it an appealing option for fusing medical images to effectively diagnose tumors and plan treatment accordingly.