Medical Image Fusion Based on Transfer Learning Techniques and Coupled Neural P Systems

Medical image fusion is an essential task for clinical diagnosis because it allows physicians to make more accurate diagnoses. Up to now, many medical image synthesis algorithms have been proposed, but the obtained images are often of low quality and do not preserve details from the input images. This paper proposes a new algorithm that allows the creation of composite images with advantages in two aspects: good quality and information preservation. Initially, a method for image decomposition is introduced, which separates the image into five distinct components: two highly detailed components (HDCs), two low-detailed components (LDCs), and a base component (BC). This methodology is founded on the utilization of the Gaussian filter (GF) and the rolling guidance filter (RGF). Then, a modified VGG19 (called MVGG-19) network is built by a transfer learning technique for the VGG-19 network to classify four classifications (MRI, CT, PET, and SPECT). The MVGG-19 is utilized to build fusion rules for HDCs and LDCs. Ultimately, a coupled neural P system (CNPS) is utilized to create fusion rules for BCs. In order to evaluate the effectiveness of the proposed algorithm, seven advanced algorithms and eight metrics were employed. The experiments have shown that the composite image generated by the proposed algorithm exhibits a marked improvement in quality and successfully maintains a substantial amount of information from the input image.