Study Of Automatic Detection Of Thyroid Tumor In Ultrasound Images

Ultrasongraphy is one of the most frequently used methods for the early detection of thyroid tumors. It plays an important role in the clinical imaging diagnosis due to its low-cost, efficiency, non-invasive and convenience. A high performance computer-aided diagnosis (CAD) system is of great significance to further improve the accuracy of thyroid tumor detection and provide physicians with a dependable second opinion. However, because of the severe speckle noise, shadowing artifacts, the poor image contrast and irregular lesion shapes, it is challenging to build a fully automatic detection and classification system for thyroid ultrasonic images. Aim of this dissertation to achieve the computer-aided diagnosis of thyroid tumor in ultrasonic image.The first part of the thesis implements the automatic detection of region of interest of thyoid tumor. The ROI is generated automatically by the texture classification of part pixels. Based on the demised ultrasonic image, the local texture and the local gray level co-occurrence matrix are firstly incorporated to measure each non-overlapping image grid. Then by taking the selected feature vector as the input of a self-organizing map (SOM) neural network, the candidate ROI and the background can be differentiated. Lastly the final ROI is obtained by removing the influence of fake ROIs, the lesions which have similar characteristics.The second part of the thesis realizes the automatic segmentation of thyroid tumor edge. Firstly, the choice based on graph theory Normalized Cut (Ncut) method partitions the original ultrasound image, and further obtains the initial contour of the tumor by employing the different gray values and spatial distributions of each cluster. Then, for a small proportion of inaccurate segmentation, a modified active contour model and manual fine-tuning mechanism is used to adjust the initial boundary to obtain the final result. Compared with the traditional edge detection methods, Ncut can accomplish the boundary extraction of tumors more accurate and automatically.The third part of the thesis achieves the benign and malignancy classification of thyroid tumor. The textures and morphologic are defined for each thyroid tumor, and Affinity Propagation (AP) clustering is used to fulfill the benign and malignancy classification. Additional, the discriminant performances of the AP clustering compares with the Fisher Linear Discriminate (FLD), the support vector machine (SVM), and the back propagation (BP) artificial neural network.Demonstration of the process of automatic detection of thyroid cancer is given in the fourth part of thesis by GUI on various parts of the visual display of test results.