Research On Algorithm Of Benign And Malignant Detection Of Thyroid Nodules Based On Deep Learning

Thyroid nodules(TNS)are clumps that appear after abnormal proliferation of thyroid cells,with an average incidence of 19% to 46%.It is one of the most common nodular lesions in the population.In the past 30 years,the incidence of thyroid cancer has increased 2.4 times,which seriously harms the health of the general public.Ultrasound imaging is a common method for its examination.In order to assist doctors in diagnosis,many studies have applied machine learning algorithms to ultrasound images.Currently,a comprehensive computer aided diagnosis(CAD)system based on ultrasound images of thyroid nodules has been implemented.With the development of CT,enhanced CT is widely used in the diagnosis of thyroid nodules.Due to the lack of high-quality public thyroid CT image data sets,there is no research to apply deep learning to the field of CT images of thyroid nodules.However,due to the artifacts and high complexity of thyroid CT images,traditional machine learning algorithms are difficult to apply to benign and malignant thyroid nodule detection tasks.In order to solve the above problems,this paper applies deep learning to the field of thyroid CT images,and proposes an algorithm based on complete thyroid CT images detection and benign and malignant identification.The main research results can be summarized as follows:1.Aiming at the problem that no research has applied deep learning in the field of CT images of thyroid nodules,a thyroid nodule detection algorithm based on deep convolutional neural networks is proposed.The algorithm uses two networks with different structures to judge the same CT image and fuses the results of the two.After simulation experiments,the final accuracy rate is 91.60%.2.Dense U-Net is proposed,which is an improved U-Net inspired by Dense Net and Res Net,and named Dense U-Net,which is used to learn semantic segmentation graphs from real data.By introducing the dense connection mechanism and using the residual module to replace skip-connect,the segmentation accuracy and training efficiency of U-Net are effectively improved.3.Aiming at the existing benign and malignant detection algorithms based on thyroid nodule ultrasound images that cannot be applied to CT images,a cascaded and multi-task convolutional neural network-based benign and malignant detection algorithm is proposed.Its algorithm is a hybrid model consisting of three different deep convolutional neural networks(CNN).First,this study use Dense U-Net to generate a thyroid semantic segmentation map.Then,the image of the thyroid region of interest of the enhanced CT is generated using the semantic segmentation map.Finally,CNNF fused with two structures is used to automatically detect the benign and malignant thyroid nodules from the original CT image and the thyroid region of interest image.Experimental results show that the improved Dense U-Net performs better than U-Net,and the dice coefficient increases from 0.945 to 0.955.In addition,the algorithm is very effective in detecting thyroid nodules.The area under the receiver operating characteristic curve and the accuracy rate are 98.49% and 95.73%,respectively.