Computer-aided Diagnosis Of Ultrasound Hyperparathyroidism Based On Transfer Learning

Hyperparathyroidism is a disease caused by too much parathyroid hormone secreted in the body.When the parathyroid gland is abnormal,it will have a greater impact on the patient’s health.It may also cause frequent fractures,bone pain,hypercalcemia and other diseases,and even threaten the life of the patient in severe cases.B-mode ultrasound examination of the neck is safe,convenient,easy to implement,and easy to repeat.It is the primary examination method for the current hospital to check for hyperparathyroidism.Because the location of the lesions found in hyperparathyroidism is not fixed,and the pathological characteristics are different,the diagnosis,treatment and positioning of hyperparathyroidism are difficult for sonographers with little experience,and it is easy to cause missed diagnosis and misdiagnosis.In this project,the target detection algorithm of deep learning requires a large number of data sets,and a large number of B-ultrasound images of parathyroid glands in different periods have been collected,and a standard data set for hyperparathyroidism detection has been established.This topic uses the transfer learning method.On the established data set,we use a variety of target detection algorithms to detect the hyperparathyroid glands,and all have obtained good detection results.Aiming at the problem that hyperparathyroidism is close to thyroid nodules and is easy to be confused,the first version of the data set is improved,a certain proportion of thyroid nodule images are added as negative samples for detection,and the improved data set is used to detect network is trained and tested.After experimental testing,the average accuracy of detection and location of hyperparathyroidism is 90.5%.Compared with the first version of the data set,the detection accuracy has been significantly improved.It further proves that the hyperparathyroidism detection network based on migration learning can effectively realize the detection and localization of hyperparathyroidism.Aiming at the difficulty of portability and operability of deep learning programs,this topic uses Java Web technology to encapsulate the proposed target detection algorithm,which allows users to remotely access through web pages and quickly and conveniently use deep learning to detect parathyroid glands.Hyperfunction,so as to realize the remote detection of hyperparathyroidism.