The Research Of Automatic Classification Of Ultrasonic Thyroid Nodules

In recent years, thyroid cancer has become one of the diseases endangering human health, and the proportion of the thyroid cancer is rising. Medical ultrasonic imaging technology has become an important means in diagnosing thyroid cancer because it is safe, effective, portative and low cost, etc. Digital image processing and artificial intelligence technology are used to process thyroid ultrasound images. They play an important role in determining benign and malignant tumor and provide doctors with auxiliary diagnostic information. This thesis mainly aims at researching the classification of tumors, which includes the features extraction and classification of the segmented images with some error.The size, shape and position of the ultrasonic thyroid nodules in different people are different due to the thyroid ultrasound image with low contrast and resolution, and mixed speckle noise. These are the main factors which influence the tumor feature extraction and region segmentation effect. In order to solve this problem, the segmentation method based on active contour models is used in this thesis. In this method, the original images are firstly preprocessed, and then the preprocessed images are segmented by the active contour model. This algorithm can segment the images quickly, but there exists some error in the segmented images compared with the original images. However, it can meet the accuracy requirement of the feature extraction and the classification.In this thesis, the segmented images are divided into several grids, and then the features of each grid are extracted to be the texture feature of the whole nodule area. Finally, the property of the nodule is judged by the percentage of the malignant in all grids. There exists a certain error between the segmented area and the actual area, which may influent the feature extraction of some sub ranges. The property of the nodules is judged by all the sub ranges’property, so the segmentation method may influence only a small fraction of the sub ranges’property judgment. In addition, the limitations of the global feature are avoided by extracting the local texture feature from the local area. The experimental results show that this method has a higher accuracy. It has a supplementary role in ultrasonic diagnosis of thyroid cancer.