Image Inpainting Algorithm Of Thyroid Ultrasound Images Based On GAN

With the rapid development of artificial intelligence and deep learning,the combination of medical images and deep learning has become the development trend of intelligent medical diagnosis.Thyroid ultrasound imaging is an important basis for the diagnosis of thyroid nodule-related diseases.However,existing images usually contain special cross markers for position of nodule,which interferes with the features and diagnostic results of the deep learning algorithm,so it is important to repair the thyroid ultrasound image with special cross mark.The image inpainting algorithm has achieved good results in the restoration of natural images.However,due to the large difference between the thyroid ultrasound images and the natural images,the application of the thyroid ultrasound images directly may cause artifacts.Therefore,the main researchs of this paper are as follows:(1)For the case that the result is prone to artifacts in the repaired areas,this paper adopts the blind inpainting algorithm,which does not need to mask the backfilling of the missing area to ensure the consistency of the repaired area and the whole image in the texture details.(2)The generator adopts the Laplacian pyramid structure,and gradually generates high-resolution images from low-resolution images,and improves the texture details of the images.In the content loss,different levels of pyramid generation iamges are added to the training loss for joint training.(3)The full-image structure enhancement、random block structure enhancement and edge-based structural enhancement constraints allow the image to focus on the structural information of the image during reconstruction.The experimental results show that the algorithm proposed in this paper has a significant improvement in visual effects compared with other algorithms,and the evaluation indicators in quantitative analysis have also improved.