Abstract: Ultrasound is the first choice of imaging modality for thyroid examination.Clinical analysis of thyroid ultrasonography is based on quantitatively evaluating the ultrasound image features in the thyroid imaging reporting and data system(TI-RADS).However,the results of quantified features are influenced by doctors’ experience level,status and other related factors.Computer-aided analysis can objectively analyze ultrasound imaging features and reduce the influence of subjective factors on the diagnostic results.But most of the existing systems are based on classic image texture features,which are abstract and absence of explicit meaning,so they are difficult in clinical using.Sonographic features of thyroid nodules which are involved in TI-RADS were extracted and quantified.Based on doctors’ clinical experience,the visual characteristics of the corresponding quantization methods were designed,which provide a basis of standardized description of thyroid ultrasound images.Statistical learning methods were adopted to establish a model of identifying the benign and the malignant thyroid nodules based on these characteristics,which provides reference recommendations for clinical diagnosis.The recognition accuracy of the model reaches 100%.

Key words: Image recognition,TI-RADS,Feature extraction,Thyroid ultrasound image