| With the development of medical imaging equipment,medical image analysis plays an increasingly important role in disease diagnosis.For example,the disease diagnosis report given by imaging equipment will play a very important auxiliary role in the formulation of surgical plans and drug treatment.Medical image segmentation is one of the most basic and important tasks in medical image processing.The accurate segmentation of the target lesion area is the key for subsequent image processing.Image segmentation algorithms have many forms,such as snakes,CV models,RSF models,clustering methods,etc.,which can not only perform accurate segmentation in simple composite images,but also in the field of medical images.Especially for the brain MR image,cardiovascular image and other aspects of the lesion area,these models can achieve good results.The classic CV model and RSF model are both active contour models based on the evolution of the level set function.Their advantages are that they can handle topological changes such as breaking and stretching of the curve in the evolution process,but their disadvantages are also obvious.This type of model is more sensitive to the selection of parameters and the position of the initial contour,and they often require a lot of tuning experiments to get a better segmentation effect.Other image segmentation algorithms,such as clustering method and graph cut method,can also accurately segment some medical images,but most methods have limitations in specific application areas.With the update of computing equipment and the increase of computer storage,deep learning methods in the era of big data have become hot research content.However,in view of the small amount of medical image data and the cumbersome problem of manually labeling training data,the neural network model is not easy to achieve satisfactory segmentation results.In order to further improve the stability and segmentation efficiency of the active contour model,this paper proposes a general image segmentation model based on the prior information of various shapes(DPLSF model).The DPLSF model is a general expression form of an image segmentation model.In the specific application process,after selecting the type of energy function and the acquisition method of prior information,the DPLSF model will have its specific expression form.Firstly,this article introduces an image segmentation model(FLSF model)based on the combination of fuzzy clustering and level set methods.By introducing dynamic update items into the model,the segmentation results of fuzzy clustering are used to restrict the evolution of the level set function,which can effectively avoid the adjustment of initial contour to obtain accurate and automatic segmentation results.Finally,an image segmentation model combining K-nearest neighbor algorithm and level set(KLSF model)is introduced.The K-nearest neighbor algorithm generates a classification criterion based on the training data,and generates a score probability matrix for each test sample which determine the category of the test sample.Different from the boundary detection function used by the CV model or the RSF model,the KLSF model uses the score probability of each pixel as the numerical result of edge detection for each pixel.The FLSF model and the KLSF model are two independent and specific application examples derived from the DPLSF model.The above three models can use the split Bregman method to minimize the energy function to speed up the segmentation.Experimental results show that the DPLSF model can not only successfully segment medical images,but also part of natural images.The quantitative and qualitative experimental results show the good segmentation effect of the model in this paper.Compared with other basic models and deep learning models,the evaluation indicators such as DICE value and F-measure have a certain degree of improvement.At the same time,the test of initial contours and parameter analysis also show that the DPLSF model has strong robustness.In the future,the three models can continue to explore the segmentation methods of 2D color images and 3D images,in order to apply to more application scenarios. |
