Feature Extraction And Visualization Of Ultrasonic Image Based On TI-RADS

In medical field,nodule is the uncertain benign and malignant tumor without diagnosis.Thyroid cancer is a malignant nodule,benign nodules mostly are inflammatory nodules or cysts.Due to most thyroid nodules are asymptomatic,the disease is easy to be ignored.At present,the commonly used diagnostic method for thyroid diseases is ultrasound diagnosis,but there are still some subjective defects in the ultrasonic diagnosis.With the wide application of medical imaging equipment and the rapid development of digital image processing technology,more and more researches are made on computer aided diagnosis using image processing.The main purpose of computer aided diagnosis is to classify the ultrasonic images accurately by computer recognition,and to provide the reference results for doctors and patients.The goal of this thesis is to study the feature extraction and data visualization of thyroid nodule ultrasound image based on TI-RADS table.The purpose of this study is to study the ultrasound images of thyroid nodules with different grades of TI-RADS,and to classify the results of the nodules and the differences of the characteristics of different grades.The main contents of this thesis include three parts:pretreatment of thyroid nodule ultrasound image,feature extraction of thyroid nodule ultrasound image and visualization of thyroid nodule ultrasound image.Ultrasonic image preprocessing includes image denoising and image segmentation.In this thesis,a novel anisotropic diffusion model based on edge enhancement(EEAD)is applied for the speckle noise in thyroid images.According to the segmentation of the nodule in ultrasound image,we propose new Graph Cut algorithm based on edge gradient operator and shape constraint,which is obtained by minimizing the energy function.The segmentation algorithm optimizes the segmentation results of ultrasound image are not accurate and edge impulsive phenomenon.Feature extraction method of thyroid nodules based on TI-RADS table is proposed to quantify the characteristics of ultrasound images.The characteristics of nodules are divided into 5 categories:morphology,boundary,texture,aspect ratio and calcification.Based on the morphological features,texture features and other methods of feature extraction,34 kinds of features are obtained.The correlation data,T test and Clustering algorithm are used to quantify and test the feature data.The last part in this thesis is visualization,which include cluster analysis and visual design.The clustering algorithm divides the data into different grades according to the sample rule of the feature,then the classification results are displayed by the visual layouts.In the experiment,the ant colony algorithm based on genetic algorithm(GACO)used to cluster single feature,multi view weighted clustering(TW-Kmeans)for multi class feature clustering.The visualization experiment designs different visual layouts for different data structures,the visualization includes the basic information part and the classification part of the nodules.We use the circular partition graph and the rectangular tree graph to represent the basic information.The results of the classification of the nodules and the differences of the characteristics of different levels are shown by the radar map,parallel coordinate and the star scatter plot.