Performance analysis of multimodal medical image fusion using AMT-DWT-based pre-processing and customized CNN for denoising

Multimodal medical image fusion involves integrating information from various modality source images to create a fused image, facilitating straightforward and reliable diagnosis. In this paper, we present a novel approach utilizing an adaptive multilevel thresholding (AMT)- discrete wavelet transform (DWT)-based pre-processing method, leveraging the strengths of multiple image enhancement techniques through averaging and fusion in the frequency domain via DWT, which enhances the quality of the source image. The proposed enhancement technique is compared with several conventional image enhancement methods for performance validation. Thereafter, the enhanced images from different modalities are fused using an appropriate fusion rule such as conventional PCA or the Mean-Max fusion rule. To further de-noise and improve the visual quality of the fused image, a custom CNN incorporating two proposed activation functions is employed. This study aims to assess the performance of the mentioned fusion methods with the proposed activation functions. The proposed framework exhibits superior performance in terms of fusion metrics compared to many state-of-the-art image fusion models.