Image Texture Analysis And Haptic Display For Medical Diagnosis

Computer-aided diagnosis(CAD),as a new diagnostic technique,can help doctors understand the patient's condition by efficient and objective diagnosis.In order to better realize CAD,this paper conducted related research from two aspects of visual perception and haptic perception.In visual perception,it mainly analyzed the texture features of the 2D and 3D medical images and gave the diagnosis results according to the analysis results.In haptic perception,it mainly focused on the texture haptic rendering of the image and displayed surface haptic information of the virtual object through the interactive device.The main researches of this paper are as follows:(1)In the process of 2D medical image recognition,aiming at two applications in the melanoma skin disease and the liver tumor,this paper proposed an automatic diagnostic framework based on CNN.The framework mainly includes three parts: image pre-processing,segmentation and classification.According to different applications,the corresponding processing methods are adopted for these three parts.In the recognition of the melanoma,the illumination evaluation was added in the pre-processing step for relieving the illumination variation problem,and the Grabcut segmentation method based on color difference was proposed by means of the color difference between melanoma and surrounding normal skin.In the recognition of the liver tumor,median filtering was adopted in the pre-processing for the occurring salt noise in liver CT images.Because of having hundreds of CT images for every group of liver data,the U-Net automatic segmentation method was adopted to realize the liver segmentation efficiently.Because the 2D medical images mainly display the texture information in the lesion areas,in the classification stage,this paper used the CNN to extract the image texture features for end-to-end automatic image classification,and further realized the CAD.(2)Compared with 2D medical images,3D images can display more comprehensive information.Therefore,this paper analyzed the texture information of the 3D liver images based on the improved LBP methods and completed the 3D liver tumors recognition finally.The improved LBP methods involved mesh LBP,VLBP and LBP-TOP.In the mesh LBP,the 3D mesh model of the liver was obtained by using the surface rendering method of marching cubes,and then the LBP texture analysis method was adopted for the 3D mesh model.In the mesh LBP method,this paper proposed a new key point extraction method which made use of the regional characters of the liver tumor.The new extraction method extracted the five points with the highest shape index of the liver image.The VLBP and LBP-TOP method extend the LBP method to dynamic texture analysis,which consider the LBP features of the images of the first n frames and the next n frames.Finally,the effects of the improved LBP methods in the 3D liver tumor recognition were demonstrated by experiments.(3)Haptic perception is an important source of information in addition to visual perception.Based on this,this paper proposed an enhanced texture force calculation method to help doctors further understand and analyze the patient's organs.This paper added new feedback force based on the traditional texture force as the gradient value of the pixel,which better displays the surface information of the object.The new feedback force can exist when the gradient value reaches the set threshold.Finally,experiments demonstrated that the enhanced texture force can improve the accuracy of image recognition and make the operator have better immersion and realism compared with traditional texture force.