Classification Of Thyroid Nodules Based On Multi-scale Feature Fusion

The incidence of thyroid cancer has been increasing year by year,ranking the fourth place among women in the world.Early detection of thyroid cancer with surgical treatment,the five-year survival rate of patients up to 96 %.Therefore,early detection and diagnosis is the key to control the disease.Ultrasound has the advantages of nontrauma,real-time,low price and simple operation,and is the most commonly medical imaging for early detection and diagnosis.However,using ultrasound to accurately diagnose thyroid cancer requires radiologists with clinical experience.With the rapid development of deep learning,the application of deep learning to deal with medical image problems has become a trend.Using deep learning to accurately and rapidly diagnose thyroid cancers has become a research hot spot.This paper mainly discusses the thyroid nodule classification based on convolutional neural network.This paper first reviews the development and research status of thyroid nodule classification.Then,the research history and significance of deep convolutional neural network are introduced,and its basic principles,including its structure,characteristics and optimization methods,are emphatically introduced.The main contributions of this paper are as follows:In view of the lack of available public data sets for researchers in thyroid nodule ultrasound image,the methods of collecting,cleaning and screening high-quality thyroid nodule images are introduced.The application of multi-scale feature fusion in thyroid ultrasound image classification was discussed.Two kinds of multi-scale fusion modules are designed based on residual network.Aiming at the shortcomings of ignoring the high resolution features,a fusion mechanism of high resolution channels and low resolution channels is designed based on the residual network structure.