Pulmonary Nodule Detection And Identification Based On CT Images

Pulmonary cancer is the most common primary malignant tumor. In recent years, pulmonary cancer Morbidity and mortality increased rapidly and has been the highest morbidity and mortality of cancer. The performance of early pulmonary cancer is the nodules on pulmonary CT images. CT is an important means of detecting pulmonary nodules, it can be used to detect malignant nodules in early lesions and greatly improve the survival rate of pulmonary cancer patients. With the development of CT, especially the emergence of MSCT, the workload of radiologist has greatly increased, resulting in missed rate and false detection rate increased. Computer-aided diagnosis system(CAD) can help radiologists to detect and diagnose and to Improve the detection efficiency and detection accuracy.In this paper, applying to the difficulty of segmentation and detection on lung CT image and medical characteristics of lung CT image, a set of automatic detection and recognition algorithms is presented on pulmonary parenchyma segmentation, suspected nodule analysis and extraction, and pulmonary nodule detection and Identification. Using the optimal threshold method to segment pulmonary parenchyma based on Lung CT images of the gray feature and to obtain a complete lung parenchyma. Using the threshold segmentation of moment-preserving principle and multi-scale enhancement filter of Hessian matrix to extract the suspected lung nodule, to enhance the circular suspected pulmonary nodule and suppress trachea, blood vessels and other liner interference areas. In order to reduce the false positive rate, This paper proposed a pulmonary nodule detection and Identification base on Simulated Annealing(SA) Algorithms and Fisher classifier. Using improved SA to get the best feature combination, using the best feature combination to establish a classifier based on Fisher discriminent to suppress endpoint of vessel and the cross point of vessels by reducing false positive and retaining high true positive. Experimental results show that this paper proposed the algorithm has good performance on the sensitivity and specificity.