Research Of Cell Image Detection And Segmentation Based On AlexNet And Level Set Method

Cell image detection and segmentation are the fundamental of medical image processing.Cell images have a variety of features,uneven staining,and the formation of adhesions between cells.The existing image detection and segmentation algorithms have unsatisfactory effects on the detection and segmentation of cell images with blurred boundaries and low contrast.Deep learning is currently a hot topic in machine learning research.Convolutional neural network has relatively few applications in pathological cell image detection.However,it has obvious advantages in image processing applications.The level set method is a classical algorithm for image segmentation,and it still has much room for development in cell image segmentation tasks.This paper applies the AlexNet convolutional neural network to cell image detection tasks and level set method to cell image segmentation tasks.The main research work of this paper is as follows:(1)Design the cell image detection model.The content of cell images is disordered with many goals and complex backgrounds.In order to effectively extract multi-dimensional image features of complex pathological cell images,the structure of AlexNet convolutional neural network is improved.The proposed detection model has significantly improved the detection performance on breast pathological cell images and its accuracy reaches 84.89%.(2)Design the cell image segmentation model.Taking into account the characteristics of the breast pathological cell image and the situation in which some cells may overlap or deform.This paper establishes a new energy function for the level set method,which combines image region information with edge information and cell shape prior information.Our proposed segmentation model performs well in breast pathological cell image segmentation tasks and accuracy reaches 88.93%.(3)Combine cell image detection model and segmentation model.Taking into account the characteristics of the cell image features,the detection method and segmentation method,the outputs of the detection model are used as the initialization value of the segmentation model so that a complete detection and segmentation model is constructed.This model is more suitable for breast pathological cell images on which it performs well in detection and segmentation tasks.