Research On Medical Image Segmentation Model Based On Atlas Correction And Feature Fusion

Medical image segmentation has extensive application and research value in medical research,clinical diagnosis,pathological analysis,surgical planning,image information processing,computer aided technology and other medical research and practice fields.At present,most state-of-the-art image segmentation algorithms can achieve excellent segmentation performance and accuracy under certain conditions,but it is difficult to overcome the sensitivity of the model itself to initial conditions,parameter settings and image noise.Therefore,this paper proposes some innovative and progressive models for the shortcomings of these methods.In the region-based segmentation algorithm,the Region-Scalable Fitting model can effectively segment gray-scale uneven images,but exhibits sensitivity to initial contours and parameters.In the cluster-based segmentation algorithm,the Coherent Local Intensity Clustering model can simultaneously segment the image and correct the bias magnetic field,but it shows sensitivity to image noise.Therefore,this paper proposes a new segmentation model with atlas correction term,which fully takes into account the advantages and disadvantages of regional intensity fitting and local intensity clustering.Specifically,the new model defines a new atlas energy term,which is transformed from the segmentation result of the Coherent Local Intensity Clustering model.We add the newly defined atlas energy term to the energy functional of the existing Region-Scalable Fitting model,and we get the energy functional of the new model.The addition of the atlas energy term can constrain and correct the evolution of the level set function,which not only accelerates the speed of the energy functional minimization,but also ensures the accuracy of the segmentation result.The segmentation results in the experiment not only qualitatively analyze the segmentation performance of the existing model and the new model,but also use some common similarity measures to quantitatively analyze the segmentation accuracy.This shows that our model has a greater degree of improvement,both in terms of segmentation accuracy and robustness to initial conditions,parameters and image noise.For the segmentation algorithm based on atlas fusion,whether it is the most basic majority voting method in the global strategy or the weighted voting method widely used in the local strategy,both of them only surround the measurement between the atlas image and the target image or the atlas label result,without mining and using the information of the target image itself.In particular,the error caused by the image noise to the registration result is transmitted to the label selection and fusion process,so that the accuracy and reliability of the estimated segmentation result are greatly reduced.Therefore,this paper proposes a atlas fusion model based on the intensity features of the target image for image segmentation.Specifically,we introduce a feature information based on the difference in image intensity.This feature records the intensity variation characteristics of the target pixel in each direction to predict theprobability that the pixel is a balanced or unbalanced point.The detection of balance points and unbalanced points can be used to eliminate the effects of noise points present in the image.Further,based on the performance level estimation mechanism based on the selection and iterative methods,the new model combines the global weight calculated by similarity measure between the spectral label result and the predicted segmentation result and the local weight calculated by the probability feature derived from the image intensity difference,and then the final atlas label fusion is performed.The segmentation results in the experiment are qualitatively analyzed on the segmentation performance of the existing model and the new model,and some common evaluation criteria are used to quantitatively analyze the segmentation accuracy to illustrate our new construction.The model has made some breakthroughs,both in terms of segmentation accuracy and robustness to registration results,number of atlas,and image noise.