Thyroid diseases include diffuse thyroid disease and nodular thyroid diseases,both of them are harmful.Ultrasound examination is a commonly used method for diagnosis of thyroid diseases.Doctors provide a diagnostic result based on their knowledge and experience by observing the characteristics of ultrasound images.However,different doctors possess different experiences and backgrounds,the diagnostic results are not objective and consistency.Thus,an intelligent diagnostic technique,which can provide reliable diagnostic opinions of thyroid diseases based on ultrasound images,is needed to assist doctors in clinical diagnosis.This paper implement a thyroid disease diagnosis system based on the research of ultrasound images of different thyroid diseases,the research includes the following aspects:First of all,the region of interest(ROI),prepocess and feature extraction of ultrasound thyroid images are conducted.Multi-level Wavelet Multi-sub-bands Co-occurrence Matrix(MWMCM),fibrous variant texture and the longest highlighting run-length based on GLRL are proposed based on diffuse disease,while eccentricity and compactness of the nodule is extracted based on thyroid nodules.Second,for diffuse thyroid disease,a two-class classicication method is proposed,and the mRMR method is applied for feature selection,combining with SVM classifier,the classification accuracy of normal,Graves’ disease and Hashimoto’s disease are 88.33%,86.67% and 93.33%.And several comparison experiments are carried out to prove the reliability of the proposed method.Third,for thyroid nodules,a classification accuracy of 85.59% and 88.20% for benign and malignant nodules are realized by training VGGNet.An image segmentation method is performed based on U-Net,and artificial features are extracted.Fusion of CNN features and artificial features rasied classification accuracy of nodules to 90.83% and 91.57%,which indicates the effectiveness of artificial features.At last,based on the research of thyroid diseases,an intelligent diagnostic system of thyroid diseases is realized to assist doctors in clinical diagnosis. |