Integrated fusion framework using hybrid domain and deep neural network for multimodal medical images

Nowadays medical images are captured through various imaging modalities for clinical diagnosis. It is more complicated to process the images modality wise to obtain feature information for better diagnosis, but it can be achieved by fusing the information together for accurate and efficient diagnosis. Image fusion is the process of gathering all the important information from two or more source images together. The resultant fused image would give more detailed information for further processing. Magnetic Resonance (MR) images contain clear information about soft tissues and have good spatial resolution. Computed Tomography (CT) scan is good to be in three-dimensional imaging with less scan time and high-resolution images. In the present paper, it is proposed to develop an algorithm to fuse CT and MR brain images using hybrid technique focusing primarily on healthcare. The hybrid algorithm is developed using Discrete Wavelet Transform (DWT) and Deep Neural Network to have both spatial and spectral domain features together to offer more accuracy compared to the existing fusion algorithms. The evaluation metrics show better performance of the proposed fusion method compared to wavelet transform based fusion. The average entropy and standard deviation are 7.41 and 84.18 respectively for the proposed DWT-CNN fusion, representing a 5.58% and 26.74% of improvement compared to wavelet transform based fusion.

Automated summary

Nowadays, medical image fusion plays an important role in clinical diagnosis, allowing for the integration of multiple imaging modalities to obtain more accurate and efficient diagnostic features. In this study, a hybrid algorithm combining Discrete Wavelet Transform (DWT) and Deep Neural Network is proposed to fuse CT and MR brain images, with a focus on healthcare. Evaluation metrics demonstrate that the proposed fusion method outperforms traditional wavelet transform-based fusion, with 5.58% and 26.74% improvements in average entropy and standard deviation, respectively.