Research On Automatic Identification Of Fungi And Classification Of Leucorrhea Cleanliness Under Leucorrhea Microscopic Images

Leucorrhea microscopic examination is one of the routine clinical exams in hospital,as it provides important information for evaluating gynecological diseases.Traditional manual microscopy is adopted by most hospitals for leucorrhea microscopic examination;however,it has both a high false-positive rate and low efficiency.Because of the complex background of the leucorrhea microscopic image,the medical staff in the laboratory can easily produce visual fatigue under high intensity and long time working.Nowadays,with the combination of digital image processing and medical examination,the technology of automatic recognition and classification of biomedical images has made rapid progress.It has become the development trend of modern medical inspection that the machine vision technology is used to share part of the inspection tasks and even replace the work of inspectors.Through automatic identification of the fungi and trichomonads,and the automatic classification of vaginal cleanliness,the doctors can diagnose three kinds of gynecological diseases,such as colpitis mycotica,trichomoniasis vaginitis and bacterial vaginitis.In this thesis,through microscope optical imaging system and auto focus module,clear microscopic images of eucorrhea could be acquired.Based on the theory of cell morphology,digital image processing and artificial neural network theory,this thesis has completed the identification of the fungi and trichomonads and the classification of leucorrhea cleanliness.The main works are as follows:Firstly,a series of image preprocessing operations has been carried out on the microscopic image,including image gray scale,spatial domain filtering,morphological basic operation,threshold segmentation and so on,and the foreground has been extracted from the background.Secondly,the characteristics of the connected region has been extracted based on the morphological features of the cell,and the connected region was screened.Then,the fungi have been automatically identified by template matching,cluster analysis and concave pitting detection.The recognition rate is 92.9%,the miss detection rate is 7.1% and the false positive rate is 1.9%.Thirdly,an improved Kalman filter background reconstruction method is proposed for detecting athletic trichomonads,which can extract accurate trichomonads region,suppress smear and ghost region,quickly eliminate false positives,adaptable continuous or abrupt illumination changes,the lens distance changes and the impact of the lens shift.Detection of 50 samples,shooting 500 fields of vision,each field of view to shoot 15 pictures,the false detection rate is 3% and the missed detection rate is 5%.Finally,BP neural network is used to determine the cleanliness of leucorrhea,and BP neural network is trained by elastic BP algorithm.There are four kinds of cleanliness of the microscopic image classification,each cleanliness has 250 pictures.The overall classification accuracy rate is 94.8% and the classification error rate is 5.2%.The experimental results show that the algorithm studied in this thesis can meet the requirement of testing the leucorrhea.There is no report on automatic identification of fungi and trichomonads in microscopic images of leucorrhea and the automatic classification of leucorrhea cleanliness before.