Study On Medical Image Segmentation And Classification Algorithm Based On Meta-learning

Medical image analysis can reflect the health of the human body and assist doctors in diagnosis and treatment,so it has been widely concerned by scholars.Because medical images involve patient privacy,public data sets are scarce,and the number of samples available is also small,which brings great inconvenience to the intelligent analysis of medical images.But,meta-learning has advantages in dealing with few-shot learning.Thus,this paper studies the method of medical image analysis based on meta-learning,taking the classification and segmentation of medical image as the research object.The main work of this paper is as follows:(1)In order to achieve better results in medical image segmentation,it is usually necessary to train a large amount of manually labeled data.However,there are few public manually labeled data sets and samples.This paper proposes a medical image segmentation method based on metalearning to solve the above problems.First,this paper studies and practices the meta-transfer learning algorithm,and then combines the meta-transfer learning with U-Net medical image segmentation,so that U-Net medical image segmentation can be applied to few-shot learning.Then,the method is extended to the tuberculosis X-ray medical images,using the NLMMontgomery County-Chest X-ray Database to segment the lungs in the tuberculosis X-ray medical image.Experimental results show that the segmentation effect of this paper has been further improved in eval uation indicators such as accuracy,Io U and recall.(2)Aiming at the problem of low reusability of medical image classification network model,this paper proposes a medical image classification method based on meta learning.First,because the image features in the medical image data set are not obvious,the U-Net image segmentation and the adaptive mean threshold segmentation are combined to segment the chest X-ray image,it could enlarge the differences between classes.Then,this paper analyzes and studies the Model-Agnostic MetaLearning(MAML).According to the features of the pre-processed chest Xray image,the network structure of MAML is improved to get more features in the medical image.At the same time,the activation function and loss function are also improved.Finally,the experimental results show that the proposed algorithm can improve the accuracy of medical image classification.