An efficient Dense-Resnet for multimodal image fusion using medical image

Nowadays, the visual content of various medical images is increased through multimodal image fusion to gather more information from medical images. The complementary information available in various modalities is merged to increase the visual content of the image for quick medical diagnosis. However, the resultant fused multi-modality images suffered from different issues, like texture distortion and gradient, mainly for the affected region. Thus, a hybrid deep learning model, Dense-ResNet is designed for the fusing of multimodal medical images from different modalities in this research work. The images from three different modalities are collected initially, which are individually pre-processed by using a median filter. The pre-processed image of the spatial domain is transformed into a spectral domain by applying Dual-Tree Complex Wavelet Transform (DTCWT), and the transformed image is segmented using Edge-Attention Guidance Network (ET-Net). Finally, the multimodal fusion of medical images is performed on the segmented images using the designed Dense-ResNet model. Moreover, the superiority of the designed model is validated, which shows that the designed Dense-ResNet model outperforms as compared with other existing multimodal medical image fusion approaches. The Dense-ResNet model achieved 0.402 Mean Square Error (MSE), 0.634 Root Mean Square Error (RMSE), and 47.136 dB Peak Signal to Noise Ratio (PSNR).