Research On Key Techniques Of Early Warning For Continuous Circular Capsulorhexis Surgery

With the continuous development of machine vision and imaging technology and its application in computer-aided diagnosis,using computer technology to assist surgeon in accurately performing cataract surgery has significant clinical importance.Currently,Continuous Circular Capsulorhexis(CCC)in cataract phacoemulsification surgery mainly relies on the experience of surgeon,which poses a great challenge for young surgeon with limited experience.Moreover,traditional CCC surgery warning schemes mostly rely on additional dyes or equipment,not only increasing costs but also adding extra steps to the surgery,potentially increasing the risk of surgical complications.This article proposes a warning scheme for the CCC process in cataract phacoemulsification surgery,which consists of three steps: firstly,locating and tracking the position of the capsulorhexis forceps during phacoemulsification surgery;secondly,extracting the capsular edge in real-time during the CCC surgery operation and generating a standard virtual capsular boundary;and thirdly,combining the position of the capsulorhexis forceps and the virtual capsular boundary to develop a warning scheme for the continuous circular capsulorhexis surgery operation.The specific research contents of this article are as follows:First,for the localization and tracking of instruments in phacoemulsification surgery,a method based on kernel correlation filtering and another method based on convolutional neural networks are proposed.The kernel correlation filtering method utilizes the first frame image target and kernel method,convolving the improved HOG(Histogram of Gradient)feature of the target with a kernel function to construct a specific response filter.During the tracking process,the position of the target area is determined by calculating the similarity between the target area and the filter,and the filter is updated using new samples.Second,the method for instrument localization and tracking in phacoemulsification surgery based on convolutional neural networks employs an end-to-end fully convolutional neural network for training.The network uses the first 10 layers of VGG-19 as a feature extraction network to extract feature information from the images,and utilizes a residualbased deconvolution module to abstract the features into high-resolution feature maps.Finally,the confidence heat map of the capsulorhexis forceps keypoint is output through convolution.By using this confidence heat map and combining the information between frames,the coordinates of the capsulorhexis forceps keypoint is generated.Experimental results show that the average detection error in video sequences for the instrument localization and tracking method based on convolutional neural networks is 4.6526 pixels,and the tracking accuracy curve indicates that this method achieves the best results.This performance is significantly better than the kernel correlation filtering method;however,this method requires a certain number of training sets and manual annotation of keypoint to ensure effective training.Third,an effective method for extracting virtual capsulorhexis boundaries in CCC surgical operations has been proposed.This method first locates the entire eye region by analyzing the connected components of specular reflection points in cataract surgery videos.Then,a ridge edge extraction operator is designed for eye edge variation,utilizing this operator to extract pupil edge features and detect the pupil margin using the Hough transform.Finally,the scleral margin and virtual capsulorhexis boundary are calculated and generated based on the pupil margin.This method has been tested on cataract surgery videos,achieving an accuracy of 98.52% in locating specular reflection points.We compared the operator proposed in this paper with Sobel,Scharry,Laplace,and Canny operators,and the results showed that our operator has the smallest mean square error and the highest structural similarity.Forth,in order to achieve continuous circular capsulorhexis surgery operation warning,the keypoint localization method of the instruments is combined with the virtual capsulorhexis boundary method.By calculating the positional relationship between the keypoint and the virtual capsulorhexis boundary,the standard virtual capsulorhexis boundary is displayed in real-time during the surgery and warnings are issued where potential problems may arise.This paper proposes a continuous circular capsulorhexis surgery operation warning scheme,which tracks intraoperative instruments using kernel correlation filtering and convolutional neural networks.By employing the ridge edge operator to generate virtual capsulorhexis boundaries,this scheme realizes continuous circular capsulorhexis surgery operation warning,bringing significant implications for the development and advancement of modern computer-assisted surgical operations.