Automatic Classification Technology Of Optical Image Of Human Fecal Occult Blood Test Reagent Card

As a routine test item in the laboratory,stool routine examination plays an important role in clinical testing.At present,fecal occult blood testing still uses manual processing plus card reading.This method has the disadvantages of low efficiency and easy pollution.With the development of intelligent classification technology,medical detection automation has become the development trend of modern medical inspectionThis thesis studies the automatic classification technology of fecal occult blood test reagent card.Relying on the automatic manure routine analyzer,the image of the fecal occult blood test reagent card is obtained through the camera,and the image processing technology and machine learning technology are combined for the automatic classification of the reagent card.The main work of this thesis is as follows:First,the optical image of the fecal occult blood test reagent card is manually labeled,and mark it as negative,weakly positive or positive according to the light depth of the reaction line(T line)and append a score between [0,10].The image is then preprocessed.The reagent card is contaminated by feces during actual use.By comparing the results of various image segmentation techniques,this thesis uses a combination of threshold and frequency domain to segment the optical image and use the LAB color space model to assist.Verify that the reaction area in the reagent card image is accurately extracted;Secondly,the feature extraction and selection of the reaction zone of the reagent card were studied.By analyzing the image and basic characteristics of the reaction zone of the reagent card,it was concluded that the basic characteristics of the single image did not suit the classification criteria directly as the reagent card,so this thesis proposes the combination of texture features and color features serves as the basis for the classification of reagent cards.Finally,the multi-classification algorithm and regression algorithm of support vector machine are discussed.The extracted feature vectors are processed,divided into training set and test set,and the training set is sent to the support vector machine to adjust the parameters and pass the verification set.The trained model is validated to obtain the optimal classification model.The test model is used to test the obtained classification model to obtain the final experimental results.It is proposed to use the regression model to score the data to obtain more accurate classification results and assist the doctor to be more accurate that judging and mastering the actual situation of the patient.The characteristics of this study are that the optical image of the fecal occult blood test reagent card that needs automatic classification has complex background,and the reagent card is classified into a multi-classification problem.There is no obvious boundary at the classification boundary,and the image has no obvious distinguishable features due to the pollution situation.The experimental results show that the accuracy of the automatic classification of the optical image of the fecal occult blood test reagent card reaches 98.4%,and single card average detection speed of 0.69 seconds,which met the requirements of clinical testing.At present,the technology has been put into clinical use in some hospitals in China.