Research On Classification Method For Esophagus Endoscopic Ultrasonography Image Based On Graph Theory

Esophageal cancer is a common malignant tumor in digestive tract. Every year, about300,000people died from esophageal cancer around the world. So it’s a serious threat to people’s lives and health. Moreover, without obvious symptoms, early esophageal cancer affects the doctors’ diagnosis and misses the most treatable stage. The development and popularization of endoscopic ultrasonography improves the resolution of esophageal imaging and it becomes a necessary inspection means of early esophageal diseases. The diagnosis of esophageal EUS examination is affected by the subjective factors of the doctors, so its applications are restricted in different degrees. With the spreading and application of computer technology, it is of great significance to analyze medical images with computer technology and assist doctors in finding lesions.The dissertation mainly studied the classification of the EUS images, including esophageal leiomyoma, early esophageal cancer and normal esophagus. Based on graph theory, the classification of esophageal EUS images first transformed the images into graph structures. Combined with the thermodynamic depth information model, it built K-layer center expansion subgraphs around every vertex, computed their Shannon entropies and then took the entropies as texture feature vectors of the graph. Finally, depth-based graph kernel and support vector machine were used to train and classify graphs.The dissertation analyzed esophageal EUS images’ texture feature and studied pattern classification algorithm in order to obtain more accurate and objective diagnosis information. It promotes more widely use of endoscopic ultrasonography in the digestive diseases, further improves the accuracy and application value of endoscopic ultrasonography and help in-depth analysis and judgment for refractory diseases.