Thyroid Nodules Detection And Segmentation Method Research Based On Deep Learning

Thyroid is an endocrine organ that affects the metabolism of human body by secreting thyroid hormone.Thyroid nodule is considered as a major clinical manifestation of thyroid abnormalities.With the development of ultrasound diagnosis technology,more and more nodules will be detected.However,the ultrasound image itself has the characteristics of being greatly affected by noise and poor imaging quality.In addition,the nodule area and normal tissue area have the characteristics of low contrast,fuzzy boundary,high quality and so on Because of the different shape and size,it is more difficult for the film reader to diagnose.Once misdiagnosis or missed diagnosis,it will have serious consequences.Convolutional neural network has been deeply studied and applied in the field of image processing,The medical image features are extracted and learned by multilayer convolution neural network,and the corresponding diagnosis and prediction results are given,which is the application of deep learning algorithm to help doctors quickly and accurately locate the disease and assist doctors to make treatment plans.Therefore,this paper uses deep learning algorithm to detect and segment thyroid nodules in thyroid ultrasound images.Aiming at the problem of thyroid nodule detection,according to the imaging characteristics of thyroid ultrasound image,this paper uses Cascade R-CNN to detect thyroid nodules.In order to improve the detection effect,this paper improves the network on the basis of the network.The proposed method combines the deep residual network with the feature pyramid network to extract multi-scale features and detect nodules with variable shapes and sizes.In order to make the network get more nodal information in the process of feature learning,this paper inserts a non local block in the backbone network to increase the global information in the process of network learning and add more semantic information for the feature extraction layer.In order to improve the positioning accuracy of the model and accelerate the convergence of the network,this paper uses K-means clustering algorithm to cluster the nodule size in thyroid image to get the anchor frame size suitable for this paper.Then,the postprocessing algorithm is improved to further adjust the detection frame of the nodule to improve the detection ability of the model.The experimental results show that the proposed method has good detection performance in thyroid nodule detection.Aiming at the problem of thyroid nodule segmentation,this paper studies the thyroid nodule segmentation based on the results of U-Net and target detection.In order to improve the segmentation ability of the model,this paper introduces the attention gate module and batch normalization,and improves the loss function to improve the segmentation ability of the model.Finally,in order to obtain more precise segmentation results,this paper introduces the attention gate module and batch normalization,In this paper,the detected nodules are clipped and then segmented.The experimental analysis and comparison show that the proposed method can improve the segmentation accuracy.