Research On Multi-tasking Method Of Combined Ultrasound Thyroid Nodule Segmentation And Classification

Ultrasonic images have the advantages of safety,non-invasiveness,and good diagnostic performance.Therefore,ultrasound examination has become the preferred method for thyroid disease examination.The physical characteristics of thyroid nodules,such as size and shape,are an important basis for judging benign and malignant thyroid nodules in clinical medicine,and the diagnosis of benign and malignant thyroid nodules can also affect the judgment of physical characteristics such as size and shape of thyroid nodules.However,ultrasound images have problems such as background noise,low contrast,variable morphological scales of thyroid nodules,blurred nodule edges,and unbalanced classification of good and evil.Therefore,it is difficult to accurately segment and classify thyroid nodules.Aiming at the above problems,this paper proposes a joint multi-task network model for simultaneous automatic segmentation and classification of thyroid nodules.First,preprocessing operations such as enhancement,denoising,data expansion,and normalization are performed on the ultrasound image.Then design a joint multi-task network model,which uses the fully convolutional network as the backbone shared network,and shares the extracted shallow features to the multi-task branch network.In the segmentation network branch,first add the deep convolution block to obtain the deep features of the segmentation branch,and then perform operations such as upsampling on the deep features,At the same time,a multi-scale convolutional attention module M-CBAM is proposed to improve the convolutional attention module CBAM,The upsampling results are spliced with the feature tensor after each feature extraction stage of the backbone shared network after skip connections with multiscale convolutional attention modules,Recover image information resolution while obtaining segmentation deep features,Preserving Segmentation Edge Feature Profiles Using Skip Connections with Multi-Scale Convolutional Attention Modules,Enhance feature tensor correlation,reduce nodule edge blurring,and further improve segmentation performance.At the same time,the multi-scale convolutional attention module is integrated into the classification branch,and the classification performance is optimized by using the multi-scale convolutional attention module and the residual module.Finally,the joint cross-entropy loss function is used to adjust the shared parameters to ensure the convergence speed and robustness of the model and improve performance.It is verified by comparative experiments that the joint multi-task network model segmentation classification index has better segmentation and classification results than the single-task deep learning network,It can effectively deal with the problems of multi-scale thyroid nodules and fuzzy edges of nodules,and reduce the impact of unbalanced classification of good and evil,which has certain on-site application value.