Multimodal Medical Image Fusion Based On The Adaptive Iterative Least Squares Filter And Multi-scale Decomposition

Medical images play a vital role in clinical diagnosis.Different images convey different medical information.CT(Computed Tomography)is more sensitive to highdensity areas,whereas MRI(Magnetic Resonance Imaging)can show soft tissue structures more clearly.Since the information contained in single-modal medical images is limited and cannot meet the increasingly complex medical diagnosis needs,doctors need to analyze multi-modal medical images at the same time,which increases the cost of disease identification and increases the diagnostic burden on clinicians.In this paper,we propose an intelligent dual-modality medical image fusion framework for two commonly used medical images(CT and MRI),which attempts to extract effective dual-modality information into a single image through a reasonable decomposition and fusion algorithm to facilitate clinical diagnosis.In the medical image decomposition part,an iterative least squares filter(ILSF)algorithm based on global optimization is adopted,which utilizes additive semiquadratic minimization to meet real-time application requirements.At the same time,we propose an Adaptive smoothing parameter initialization method based on nonsalient gradient(A-NSG),which adaptively initializes the smoothing parameters according to different image features,thereby avoiding the inaccuracies caused by manual tuning of parameters.Finally,a multi-scale decomposition framework based on A-NSG and ILSF is presented to achieve a thorough separation of high and low frequency information in the source image.In the part of fusion rules,we propose two different fusion frameworks according to the characteristics of high-and low-frequency sub-bands.Simultaneously,the improved fusion rules are further present and the corresponding fusion results are compared.The improved fusion algorithm is divided into two parts: At the detail layer,the high-frequency sub-bands are normalized firstly;Next,we calculate the gradient maps of the high-frequency sub-bands through the discrete differential operator;Then we propose a gradient density detection algorithm based on main gradient analysis(MGA)to modify the weight map and the bilateral filter is used to normalize the weight map;Finally,the fused detail layer is obtained by the reconstruction.At the base layer,we propose a weighted modified Laplacian method(WML)to quantitatively calculate the amount of structural information in the low-frequency sub-bands of the source image,which can further construct fused base layer in a way that is more in line with human perception.The algorithm proposed in this paper performs well in dual-modal medical image fusion.The fusion result has rich texture details and contrast,which meets the visualization needs of doctors.Compared with 9 classic medical image fusion algorithms,the proposed algorithm achieves excellent performance in both subjective and objective evaluation.The main contributions can be summarized as follows:1)A multi-scale decomposition framework based on A-NSG and ILSF is proposed,which intelligently initializes smoothing parameters according to image features.The proposed framework makes full use of high-and low-frequency features,which avoids information loss in the decomposition process and reduces the time consumption caused by manual parameter adjustment.2)A gradient density detection algorithm MGA based on major gradient analysis is proposed,which is used in detail level fusion.The gradient dense area in the detail layer can be detected and redundant texture information can be filtered by MGA,which corrects the weight of high-frequency sub-bands and retains much more effective textures.3)A weighted modified Laplacian method-based structural information quantization method WML is proposed,which is used for base layer fusion.Through WML,the structural information in the source image can be accurately quantified,which amplifies the difference between the structural and the non-structural area and the loss of soft tissue contrast is avoided.