Research Of Lung Nodules Detection Algorithms Based On Three-dimensional CT Images And System Implemention For Computer-Aided Detection And Diagnosis

According to the report of World Health Organization (WHO), death rate caused by lung cancer has already rised 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 in5years would be raised from14%to49%. With the advent of the low dose multislices CT, the image data which were handled by the physician drastically increased, therefore increase the pretermission or inaccuracy rate of diagnosis. The goal of the thesis is to develop a computer-aided detection and diagnosis system for lung cancer based on the CT images, which can identify the potential pulmonary nodules after the automatic analysis of CT images, and greatly helps doctors improve the diagnostic quality and efficiency.In medical diagnosis, the use of three-dimensional data with rich information instead of simple two-dimensional data has become the field of medical image processing trends, so the thesis using3D image data instead of2D image data to research pulmonary CT images. Three-dimensional reconstruction of medcial images is currently a hot topic problem, however reconstruction of data is very large and reconstruction of processing is very slow, the use of hardware acceleration is already the inevitable trend for three-dimentional reconstruction. The thesis use GPU to accelerate three-dimensional reconstruction in order to achieve real-time rendering purpose.This thesis has research some key algorithm about lung cancer CAD system, mainly the thesis include four aspects.(1) Based on three-dimensional CT images a simple and effective segmentation for lung parenchyma is proposed. Firstly, make use of the median filter for low-dose CT images preprocessing. Secondly, make use of the iterative threshold for CT images binarization. Thirdly, make use of three-dimensional connected label for CT images to remove background and small holes. Fourthly, make use of three-dimensional region growing for CT images to remove trachea. Fifthly, make use of the center finding method for CT images to separate left and right lung. Finally, through improved chain code repair the edge of the corroded lung, which obtain a complete lung parenchyma.(2) Firstly, analysis of nodule characteristics based on three-dimensinal space, calculating SI of the image pixel based on anti-geometric diffusion. Secondly, the thesis wipe out the potential nodules of false positive through multi-scale enhance images based on hessian matrix, which remove region of potential nodules. Thirdly, through the method of adaptive thresholding and GMG segment the potential potential nodules. Finally, extracting the feature of the nodules, the features include2D and3D aspects.(3) The thesis research on CPU-based acceleration of ray-casting algorithms, through the improvement of the traditional ray-casting algorithm. Firstly, since in the ray-tracing algorithm, the ray traversal data is more time-consuming,"SLAB"is introduced to calculate the intersection with the bounding box. Through parallel processing that getting the entry point and the leaving point of each light. Finally, the color cumulative calculation is executed, which displaying the3D image. The experimental results show that the algorithm can greatly improve the rendering speed and accuracy.(4) Development of CAD system processing platform that it can read DICOM medical image and display patient information. The system’s key technique include lung segmentation, ROI region extract, feature extraction, lung nodule detection, three-dimentional reconstruction, database and so on. Through friendly interface to meet the needs of practical application, and preserve interface include pulmonary function evaluation, three-dimentional vascular segmentation and the content-based medical image retrieval, its purpose is to upgrade the CAD system in the future.The experiments show that the lung segmentation algorithm is proposed can automatically extract quickly and accurately the lung parenchyma. The lung nodules detection algorithm has higher accuracy and sensitivity. Three-dimentional reconstruction can achieve real-time show. The CAD systerm fast computation and high detection rate shows much promise for nodules detection, which provide doctors with information useful for the diagnosis, it is showed that CAD has played a positive role in improving the accuracy of diagnosis of lung cancer, reducing misdiagnosis.