Study Of Medical Image Thresholding Segmentation And Evaluation Methods Based On Information Fusion

Image segmentation is a very classic image processing problem,which is widely used in many fields,such as computer vision,pattern recognition,artificial intelligence and medical image processing.In the field of medical research and practice,medical image segmentation technology is the basis of various medical image applications and the preprocessing operation for image analysis and image understanding.It provides the necessary prerequisites for medical image processing,such as image registration,image fusion,target recognition,etc.,and expand the breadth and depth of the medical image in clinical application.However,the medical image itself is complex and diverse,and there are blurred,uneven gray,low contrast and other shortcomings.Therefore,it is very easy to lead to boundaries misclassification between different organs or organizations,which increases difficulties for the image segmentation.Therefore,the medical image segmentation is still a challenging task.In this paper,we researched medical image segmentation from the perspective of information fusion.The details are as follows:1.Internal Generative Mechanism based multi-thresholding segmentation algorithm for medical imagesTraditional thresholding segmentation algorithms often only consider the gray information of the pixels in the image,while ignoring other information.In medical image segmentation,the end user of the segmentation result is "human",and the subsequent processing of the image needs to be decided by human’s observation and analysis.Therefore,it would be meaningless to segment the medical images if we ignore the psychological inference mechanism of Human Visual System(HVS).With the aim at solving the problem mentioned above,we adopted recent Bayesian brain and neuroscience theories——Internal Generative Mechanism(IGM),and proposed an Internal Generative Mechanism based multi-thresholding segmentation algorithm for medical images with human’s visual information and pixels’ intensity information in this paper.First,the Otsu thresholding technique is implemented on both the original image and the predicted image which is obtained by using the autoregressive(AR)model to simulate the IGM on the original image.Second,a special regrouping rule based on pixels’ intensity information is designed to regroup the misclassification pixels and improve segmentation accuracy.In the segmentation process,the proposed method takes the predicted visual information generated by the complicated Human Visual System(HVS)into account,as well as the details of the source image.The experimental results show that the segmented images provide more clear information which is more in line with the HVS,and the visual effects are much better with the increase of thresholds.2.Fusion scheme based multi-thresholding segmentation algorithm for medical imagesHow to improve the accuracy of multi-thresholding segmentation for medical images is an urgent problem to be solved.Images to be segmented may contain noises,or other artifacts,but many existing segmentation methods assume signal-independent additive Gaussian noise and hence their application leads to suboptimal and unsatisfying results.If original images are denoised before being segmented,the segmentation results can avoid the influence of noises.Nevertheless,other normal regions with high frequency may be affected by the denoising algorithm.Hence,simple preprocessing is not a good choice to address this problem.With the aim at solving the problem mentioned above,a fusion scheme based multi-thresholding segmentation algorithm with pixels’ spatial information and intensity information for medical images is proposed in this paper.As the class label of each pixel depends on the pixel’s gray level and neighbors’ labels,we first designed a novel fusion scheme based on effective neighborhood which takes both spatial and intensity information of pixels into account.Second,a detail thresholding segmentation case using the proposed fusion scheme is designed to demonstrate the effectiveness and superiority of the scheme.To accelerate segmentation,a discrete curve evolution based Otsu method is employed to segment the original image and its smoothed version to get two different segmentation maps.The experimental results show that the proposed algorithm can improve segmentation accuracy and it is robust against noises.3.A weighted-ROC graph based metric for image segmentation evaluationEvaluation of image segmentation algorithms is a crucial task in the image processing field.Nearly all the existing metrics treat each pixel equally in the process of evaluating segmentations.However,pixels may gain different importance.To overcome this problem,a new objective evaluation metric based on the weighted-ROC graph with pixels’ spatial information is proposed in this paper.Considering that pixels in different positions may gain different importance,each pixel is given a weight based on its spatial information.The ROC(receiver operating characteristic)graph with weighting strategy is constructed to evaluate the performance of segmentation algorithms quantitatively.The experimental results show that the proposed metric can apply to medical images and natural images,and it is robust to region imbalance.In addition,its evaluation results are in line with human’s subjective evaluation results.