Cell Image Segmentation And Reticulocyte Recognition Based On An Integrated Method

In recent years,with the rapid development of the theory and technology of computer vision,as well as the urgent need of medical imaging for modern technology,the use of computer for detection and processing of medical images has become a research hotspot in the field of modern engineering.Medical imaging is a very important part of modern medical detection,which is directly related to the final results of disease examination.At present,in modern medical engineering,more and more attention has been paid to the use of computer technology to assist clinicians to analyze cell images and detect lesions sites,so as to improve the speed and accuracy of medical detection.Using image processing and computer vision technology to process some medical images has very extremely research value and engineering application prospects.The cell images in this research refer to the digital image obtained by photographing the flattened cells on the glass cover with a micrograph.In the current research on cell image,there has not been a more universal algorithm to solve all cell image problems.Among many algorithms,image processing method and deep learning model are mainly used separately.However,there are many problems in the cell image data set in this research,such as severe uneven illumination,cell overlap and cell malformation,etc.,which cannot be solved by only one or two algorithms.Since the processing of the cell image involves clinical application,if the treatment of the cell cannot satisfy a certain accuracy rate,it is extremely easy for the clinician to misdiagnose the patient,and therefore it puts more stringent requirements on this research.In this paper,we propose an integrated method based on image processing,computer vision and machine learning to solve the segmentation and counting of cell images and the recognition of specific reticulocytes.Firstly,image processing technology is used to deal with some external disturbances and unavoidable noises of cell image as effectively as possible.Then,according to the problem of cell aberration in image preprocessing,feature extraction and machine learning classification algorithms are used to classify cells.The combined method recognizes it,and further effective post-processing is performed on different distorted cells to make the segmentation and counting effect of the cells better.For the recognition of reticulocytes,this paper uses the convolution neural network which is popular and effective at present,and finally achieves the segmentation of the whole cell images and the recognition of reticulocyte.If a set of system including hardware and software is formed in the later stage,it can realize the automation and intelligence of cell image counting and recognition and promote the development of biomedical field.The innovations in this article are:1)Aiming at the problem of image segmentation and recognition,this paper combines image processing,computer vision and deep learning to realize the segmentation and counting of cell images and the recognition of specific reticulocytes,and realizes the purpose of automation and intellectualization in engineering applications.2)In image preprocessing,aiming at the complex background and many interference factors of cell image,background demodulation and dual binarization methods are used to solve the internal and external interference of the image.3)For the recognition of reticulocytes,the method of feature extraction and classification cannot solve it effectively.In this paper,inspired by the existing classical deep neural network model,a new network model Retic Net is designed and applied to the recognition of reticulocytes effectively.