Classification Algorithm Research Of Medical Sperm Image Based On Deep Learning

With the continuous update and development of deep learning and image recognition algorithms,it has become more and more common to use computers to learn people's thinking ability and experience and assist people in daily work.Smart medicine is to apply more artificial intelligence technology to the diagnosis and treatment of clinical diseases.Is common in clinical medical images we analysis and judge the basis of a disease,illness will deep learning and applying the technology of image recognition in the field of medical treatment for medical image recognition and analysis of the use of high-performance computer to deal with a lot of medical images,is medical data is an important research direction,and accelerate the important branch of the construction of the intelligent medical system.According to the world health organization,7 to 11 percent of the world's population of childbearing age suffer from infertility,with at least 30 to 50 percent of the cases related to men.Semen analysis is the main method to diagnose male infertility,and sperm morphology analysis is one of the key factors to evaluate whether the diagnoser has disease.At present,the artificial evaluation of sperm has such defects as strong subjectivity,loose standards and time consumption.Therefore,it is of great clinical practical value to use image recognition algorithm and high-performance computer to evaluate sperm morphology and assist doctors in sperm morphology analysis.This paper was completed under the support of the key science and technology project of jilin provincial science and technology department and the cooperative project of the first hospital of medical university.The target detection of medical sperm image includes the extraction of the effective features of sperm head and the morphological classification of sperm head.The research focus of this paper is to use the algorithm of convolutional neural network to carry out the normal and abnormal dichotomy of medical sperm head and the target detection task of multi-classification.At first,this paper respectively to sperm YOLO v3 model and SSD model picture 2 classification and target detection,according to the characteristics of the two models and advantages,design a set of positioning before classification Y-V cascade neural network model,fixing the sperm images respectively,and the classification of operation,improve the detection accuracy of sperm binary classification task.In order to further meet the requirements of precision medicine and perform the multi-classification task of sperm morphology,this paper improves the structure of cascade model and proposes a multi-feature reuse MFRC model.The dense connection mode of dense block structure in Densenet network model is introduced into MFRC model,which has certain regularization effect,enhances the reuse of features,and improves the efficiency of network training stage.In addition,the MFRC model improves the feature utilization efficiency of each convolutional layer,improves the accuracy of medical sperm image recognition and classification,and solves the problem of model learning ability degradation.In this paper,YOLO-v3 model,SSD model,Y-V cascade model and MFRC model were used to classify sperm head images.The experimental results showed that the dichotomy accuracy of YOLO-v3 model was 73%,the dichotomy accuracy of SSD model was 78%,the dichotomy task accuracy of Y-V cascade model was 88%,and the dichotomy task accuracy of Y-V cascade model was the highest.In the multi-classification task,the multi-classification mAP value of SSD model is 0.69,the multi-classification mAP value of Y-V cascade model is 0.72,and the multi-classification mAP value of MFRC model is 0.79.Compared with Y-V cascade model,MFRC model improved the accuracy of multi-classification of sperm morphological characteristics by about 7%.The accuracy of MFRC model's multi-classification task is lower than that of binary classification task,which indicates that the multi-classification task of sperm morphology requires higher algorithm and model.The deep learning-based medical sperm image classification algorithm achieves the expected detection accuracy and rate in the bifurcated and multi-classified tasks of sperm head morphology.Under the influence of artificial experience and human misjudgment,it takes professional doctors two hours to detect 1000 images,while the MFRC model is used to detect the multi-classification morphology of 1000 medical sperm images,which only takes about 33 seconds on the computer,with the average accuracy of mAP value around 0.79.Therefore,the MFRC model can be used to assist doctors in the evaluation of sperm morphology,which greatly reduces the workload of doctors and reduces human misjudgment,and has good clinical practical value.