Study On The Recognition And Classification System Of Cervical Exfoliated Cells Based On Neural Network

Objective:The uterine cervix cancer is only second to women breast cancer in malignant tumor, however early detect and diagnosis can improve the survival rate of cancer patients. Examination of castoff cells is the most common as well as the effective method to discover earlier uterine cervix cancer.For the time being, diagnosis of cell section image is made mainly through observing with naked eye, and only a descriptive result can be obtained on the basis of the observer's personal experience. Therefore, owning to the subjective differences of doctors in personal diagnosis experience, varying degree of carefulness in their observation, and other aspects, the results of diagnosis of the same image obtained by different doctors may be different. With the development of technology of the computer, such technology as digital pattern process, the pattern-recognition and expert system,etc. Get more and more extensive application in the field of medicine, combine the pathology expert's knowledge and experience with the technology of the computer, and apply to recognition and classification of cells, can assist experts to diagnose effectively.Designed a recognition system of cyto-image analysis for cervical exfoliated cells based on neural network, which can analysis and processing the original microscopy cervical cell images. The system can extract morphology characteristic parameters of the cervical exfoliated cells, and make classification of cervical exfoliated cells.Methods:Computer-aided clinical cytology diagnosis is realized by utilizing a CCD video camera to capture cervical exfoliated cells, making use of digital image processing and pattern recognizing techniques to classify cells.Using the median filter method of the wavelet to implement the function of image denoising and improve the quality of image. Image enhancement is accomplished based on fuzzy mathematic, and it can increase contrast of image. Image segmentation is developed by genetic algorithms, which can precise determine the location and region of cell nuclear. At the same time, as mathematics morphology is used to images processing, segmentation processing of cell images are perfected. We extract parameters eigenvector such as, perimeter, area, width, height, round dim, rectangular inspection, elongation longitude of nuclear of cervical exfoliated cells. And neural network give the output results based on the extraction eigenvector.Results:The system can reduce the effect of spot, noise and other factors, improve the quality of images, and detect the periphery of images better and the region of nuclear more precisely, extract parameters eigenvector of nuclear of cervical exfoliated cells, and classify the cells. By comparing the output of the neural network with the conclusions of patho-doctors, neural networks can be right cells to identify make precise classification of cervical exfoliated cells.Conclusion:The system can accomplish the functions of aided diagnosis of cervical exfoliated cells images, and the adopted methods are feasible. It can provide more objective, accurate and reliable information of cervical exfoliated cells to patho-doctors. The image processing methods are feasible to early screening and diagnosis of cervical exfoliated cells, which can be extended to other cells' identification. It shows its significant value in clinical diagnosis and medical research at present and in the future.