Comparative Analysis of Convolutional Neural Networks and Principal Component Analysis for CT and MRI Brain Image Fusion: Enhancing Diagnostic Accuracy and Visualization

Medical image fusion is used for combining complementary information from various imaging modalities that can help in disease diagnosis and treatment planning. This research work presents a comparative image fusion algorithm between Computed Tomography (CT) and Magnetic Resonance Imaging (MRI) brain images using Principal Component Analysis (PCA) and Convolutional Neural Networks (CNN). The final fused CT-MRI images contain the CT structural details and MRI soft tissue contrast. The fusion process is improved by training the CNN to refine the PCA-fused images by learning the optimal feature extraction and combination through multiple convolutional layers. The aim of using CNN is to enhance the clarity, reduce noise, also preserve information in the fused images. The proposed work is analysed with a CT-MRI dataset obtained from Kaggle and qualitative metrics such as Structural Similarity Index (SSIM), Peak Signal-to-Noise Ratio (PSNR), entropy, and correlation coefficient. Research findings indicate that multimodal CNN image fusion considerably performs better than multimodal image fusion using PCA. Higher values of SSIM, PSNR obtained in CNN-based image fusion indicates improved structural similarity and less noise. The increased values of entropy, correlation coefficients and visual assessments also aid in favour with our study.