Research On Feature Extraction And Description Generation Algorithm Of Thyroid Nodules Based On Ultrasound Images

Thyroid nodules are common and frequently-occurring clinical diseases.In clinical practice,ultrasound imaging technology is commonly used to image thyroid nodules,and the generated ultrasound images are an important basis for doctors to diagnose and treat thyroid nodules.In the clinic,ultrasound images are usually read by well-trained experts who write text reports to describe the patient’s abnormalities and diseases.The reading of ultrasound images and the writing of ultrasound reports cost radiologists a lot of working time,and due to the limited time and energy of doctors,missed diagnosis and misdiagnosis may occur.Therefore,the automatic generation technology of medical ultrasound image reports,which is to automatically generate a paragraph of natural language describing the content of an ultrasound image,has attracted widespread attention in the field of artificial intelligence.In order to achieve this goal,based on medical ultrasound images,this article has conducted research in three aspects: thyroid nodule segmentation algorithm,thyroid nodule medical feature extraction method,and thyroid nodule image description automatic generation algorithm,including the following parts of work :First,the segmentation algorithm of thyroid nodules in ultrasound images is studied.In view of the imbalance of categories in the segmentation task of thyroid nodules,we set up a model cascade framework based on U-Net from the perspectives of adjusting the loss function and optimizing the data set to decompose the segmentation of thyroid nodules into two segments The task is divided roughly first,and then finely divided based on rough points.We built a thyroid nodule image segmentation dataset to test the effectiveness of our model.The results show that the model we built can well solve the problem of imbalance in the classification of thyroid nodules and improve the accuracy of segmentation.Secondly,the selection and extraction of medical features of thyroid nodules in medical ultrasound images.In view of the important role of ultrasound image features in judging benign and malignant thyroid nodules and the characteristics of existing data sets,the medical features of thyroid nodules that need to be described in the subject are selected.Aiming at the characteristics of medical images and the interpretability problems of deep learning,a feature extraction method combining neural network and traditional medical features is proposed.Experiments show that our method can accurately extract the relevant features of thyroid nodules.Finally,the research on the automatic generation algorithm of thyroid nodule image description based on ultrasound image.Aiming at the problem of inaccurate feature description in performing medical image description tasks using traditional image description models,this paper designs a dual-input single-output thyroid nodule image description model based on the encoder-decoder framework,which can synthesize images The gray level information and the outline information of the nodule.This paper builds a data set describing the image of thyroid nodules,and does the corresponding text preprocessing,trains our model,and uses the test set to test the model training effect.The accuracy of image description can reach 95%,which is much higher than the traditional image description model.