Research And Implementation Of Digital Pathological Image Feature Extraction Algorithm

In recent years,the development of science and technology has led to the rapid development of artificial intelligence.The application of AI technology in education,people's livelihood,medical care and other industries plays an important role in different ind ustries.The role of the computer technology in the medical industry is particularly important,especially in the COVID-19 outbreak.The computer aided diagnosis system through the interpretation of CT image fast and efficient,high interpretation accuracy and greatly reduce the workload of front line doctors,avoid a lot of manual operation,reduce the medical staff infection risk,so the computer aided diagnosis system research application has great significance.In this paper,the cervical cytological image was taken as the research object,and the pathological interpretation of computer-aided cytological image was completed through three steps of image segmentation,feature extraction and image recognition.Image segmentation is a key technology in ima ge processing.Accurate and effective segmentation results can enable doctors to timely detect and accurately interpret patients' conditions.Watershed transform is widely used in the field of image processing due to its advantages of fast computation speed and high segmentation accuracy.In this paper,watershed transform is improved by k-medoids clustering algorithm.The algorithm first converts the original RGB medical image into grayscale image,obtains the gradient image by using the multi-scale morphological gradient operator,then uses the k-medoids algorithm to cluster similar gradient images to simplify the gradient image,and then carries on the watershed transformation to the simplified gradient image to obtain the segmentation image.In order to complete the task of computer-aided diagnosis,it is necessary to extract the feature information in the image and identify the feature information to complete the interpretation.In order to better describe the correlation of pixels in different distances a nd directions,this paper selects the texture information of the image as the interpretation feature,and uses the gray co-occurrence matrix to describe the texture feature of the image.Traditional algorithm of computing in the gray level co-occurrence matrix,direction and the energy,entropy,similarity,moment of inertia equivalent as characteristic value,this algorithm can clearly describe the texture feature of image,but experiments show that different types of cervical cytology image under the algorithm characteristic value difference is small,it brings to the subsequent image recognition work great difficulties.To solve this problem,this paper proposes a calculation method of gray co-occurrence matrix based on vector fitting,which vectorizes the energy,entropy,similarity and inertia moment of the image in the direction of,and in the original direction,and takes the modulus of the sum of the four direction vectors as the final eigenvalue.In this paper,support vector machine(SVM)was used to train and test the image feature set.The results of three groups of test experiments showed that the recognition based on the improved feature parameters was more accurate than the traditional feature parameters,so that the cervical cytological image interpretation could be better completed.According to the idea of image segmentation,feature extraction and image recognition,this paper designs the cell image assistant diagnosis system in the MATLAB simulation environment.