Research On Assisted Diagnosis Of Thyroid Ultrasound Image Based On Channel Attention Mechanism

The incidence of thyroid nodular disease is increasing year by year,and it is one of the most common nodular diseases at present.It is particularly important to correctly diagnose benign and malignant thyroid ultrasound images.In clinical diagnosis,ultrasound detection has become a routine method for diagnosing thyroid nodules due to its real-time and less damaging advantages.However,the thyroid ultrasound images obtained by the current ultrasound detection have low resolution and a large amount of noise interference,and the diagnosis results mainly depend on the doctor’s experience and subjective judgment,making the diagnosis difficult and the workload heavy.Therefore,in order to reduce the workload of doctors and improve the diagnosis efficiency of thyroid nodular diseases,it is of great significance and practical value to develop a set of thyroid ultrasound image-assisted diagnosis system through artificial intelligence technology.In this paper,the deep learning model is improved based on the channel attention mechanism,and a thyroid ultrasound image recognition model with excellent performance is obtained,and based on this model,a set of thyroid ultrasound image auxiliary diagnosis system required by medical personnel in the diagnosis process is developed.The main research content and phased results of this paper are as follows:(1)Research and analysis of traditional machine learning and deep learning related technologies of thyroid ultrasound image recognition.A brief analysis of feature extraction and classification algorithms in traditional machine learning,focusing on the three convolutional network models of Goog Le Net,Res Net and Densenet,the characteristics of migration learning,and the architectural characteristics of the network model based on the channel attention mechanism used in this article.(2)The construction and preprocessing of the data set of thyroid ultrasound images.All the ultrasound images in this article are derived from real data provided by a hospital,a total of 1526 images.In addition,in order to enhance the contrast of the image,the MSR algorithm is used to improve the enhancement effect by linearly weighting the output results of multiple fixed scales,improve the brightness,and further improve the classification performance of the model.(3)Research the construction of ultrasonic image recognition model based on deep learning network,and conduct a comprehensive test and analysis of the model.In order to choose a suitable model,this paper analyzes the traditional machine learning model based on the combination of GMRF,MB-LBP and GLCM feature extraction algorithms with SVM,KNN,and RF classifiers and three deep convolutional neural networks based on Goog Le Net,Res Net and Densenet The model built by the network was tested for comparison and introduced an improved model based on the channel attention mechanism.Through the comparison experiment,it is known that the improved model based on the channel attention mechanism SE-Res Net has better performance,so the improved network model is selected SE-Res Net is used for system development.(4)Develop a thyroid ultrasound image diagnosis system based on the channel attention mechanism.Using software engineering methods to carry out feasibility analysis,demand analysis,overall design,database design and detailed design of the thyroid ultrasound image diagnosis system,and realize the development of ultrasound image upload,image assisted diagnosis,common user information import,and permission setting in the system.And display,and perform software testing on the system.The test results meet expectations and can be applied to actual clinical applications.