Medical Image Segmentation Based On Self-adaptive Weighted Active Contour Model

In the field of medicine,medical images such as Computed Tomography(CT),Magnetic Resonance Imaging(MRI)and ultrasound play an important role in the diagnosis and treatment of diseases.Accurate segmentation of medical image is the premise of further analysis of the focus area,so medical image segmentation has become the focus of the majority of researchers.Various segmentation methods are widely used in medical images,which may change clinical practice,help clinicians diagnose diseases,determine prognosis,make treatment plans,and track treatment responses.However,medical images often have poor image quality,and there are some bad factors such as noise,intensity inhomogeneity,and vague boundary,which bring great challenges to practical segmentation.Since the active contour model was introduced and applied to medical image segmentation for the first time,it has become increasingly popular as a general framework due to its superior performance in topology and smoothness.Aiming at the above problems in medical images,this paper proposes a novel Self-adaptive Weighted active contour model based on Local Intensity Difference(SWLD)to provide accurate medical image segmentation results for clinical research.Compared with the existing segmentation algorithm based on active contour model,the SWLD algorithm proposed in this paper is more robust in noise,intensity inhomogeneity and boundary blur.The innovation of the algorithm in this paper lies in the following three aspects:Firstly,a new adaptive weighted operator based on local intensity variance difference(LVD)is proposed to enable the weight coefficient of the energy inside and outside the model to be adaptively changed,thereby overcoming the limitations of previous algorithms in processing blurred images on the boundary and solving the problem that the model is sensitive to parameters.Secondly,the Local intensity mean difference(LMD)is added to the energy functional to speed up the convergence rate of the curve and improve the evolution efficiency of the curve.Thirdly,we eliminate the effects of intensity inhomogeneity and noise in the medical image by introducing a local similarity factor with a smoothing operator and two different neighborhood sizes.The work in this paper can assist clinical research,which is specifically manifested in two aspects:first,segmentation of parotid duct image.The morphological characteristics of the parotid duct are clinically relevant and can be used as an indicator of pathological processes.Accurate parotid ducting is the basic prerequisite for studying its morphological characteristics.Therefore,segmented parotid ducts have a prominent role in the diagnosis and treatment of parotid diseases.Second,segmentation of tongue cancer tumor images.Tongue cancer is a common oral cancer and has a high rate of lymph node metastasis.Accurate tongue cancer tumor segmentation results can assist doctors in pathological analysis and treatment plans.In order to prove the feasibility and robustness of the algorithm in this paper,we use Dice score and Hausdorff Distance(HD)as the criterion for segmentation accuracy,and compare the experimental results of SWLD model with CV model,LBF model,RLSF Models and other models for comparison.In the parotid duct segmentation experiment,the results of SWLD algorithm are:the mean Dice score is 91.3%and the mean HD value is 1.746 mm.In the tongue cancer tumor segmentation experiment,the mean Dice is 95.3%and the mean HD value is 3.101 mm.Experimental results show that the algorithm proposed in this paper is significantly better than existing active contour model segmentation algorithms in segmenting parotid duct and tongue cancer tumors,and can still obtain accurate segmentation results even on complex boundaries.