Research Of GGO Lung Nodules Detection Based On Ct Images And System Implemention For Computer-aided Detection And Diagnosis

Lung cancer has become one of the most harmfully malignancy to human, according to the report of World Health Organization (WHO), death rate caused by lung cancer has already risen to the highest among all cancers all over the world, if lung cancer can be diagnosed and cured at the early time, the survival rate could be improved. With the advent of low-dose multi-slices spiral CT, it has become an important tool for early diagnosis of lung cancer, however, the CT data which the radiologists need to diagnose has increased sharply. In order to reduce the workload of the doctor, help them intuitive and effective observation of lesions, reducing the chance of misdiagnosis and missed diagnosis, computer-aided detection and diagnosis (CAD) technology came into being. Study shows that CAD system plays an active role in improving the diagnostic accuracy and reducing the rate of misdiagnosis.Lung cancer in its early stage usually realized as a pulmonary nodules, pulmonary nodules can be divided into two categories:solid nodules and non-solid nodules, non-solid nodules often referred to as GGO nodules. According to research data shows that, GGO nodules are most likely to translate into a nodule of lung cancer, therefore, if successful in nodular stage is detected, the patient’s post-treatment will play a significant help, and can improve the survival rate of cancer patients10years.This thesis has research some key algorithm about lung cancer CAD system, mainly include four aspects:(1) Propose Lung parenchyma segmentation framework which is based on two-dimensional Otsu threshold method, this method can suppress the noise and improve the quality of the lung parenchyma threshold segmentation. At the same time, aiming at the defects of the calculate time, this thesis using particle swarm optimization method to optimize the time, in order to improve the algorithm quality. After that through the lung parenchyma sub-segmentation and repair, will obtain a complete lung parenchyma;(2) First of all, make use of the high-pass filter for low-dose CT images preprocessing according to the performance of GGO nodules in the image, and get lung parenchyma template image, in the step, a subtraction image is obtained from subtraction of original image and template image, converted into first-detection image. Then make use of gray level co-occurrence matrix for texture analysis, obtain second-detection image. After that, use the three-dimensional Hessian matrix to detect blood vessels, and remove corresponding regions in the second-detection image and get the GGO candidate regions. Extracting the GGO candidate regions two-dimensional and three-dimensional features, get ready for further feature classification;(3) According to the imbalance of the actual case of GGO nodules and non-GGO nodules in ROI, researching the classifier to balance data sets based on support vector machine (SVM), and at last, the classifier is assessed;(4) Realize the lung cancer CAD system. The key technologies of lung platform, lung nodules detection, three--dimensional reconstruction and database connection are put in this platform, at last, compress this CAD system and package into a installation package, make it portable to meet the needs of practical applications.Experiments show that the proposed lung segmentation algorithm can automatically obtain the pulmonary parenchyma. GGO candidate region segmentation algorithm can effectively retain the GGO nodules and reduce false positives. Unbalanced data classification can effectively improve the classification accuracy of the ROI.